[
  {
    "id": "lmarena-dataset",
    "match": [
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/document_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/image_edit/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/image_to_video/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/search_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/text_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/text_to_image/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/text_to_video/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/video_edit/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/vision_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/6931a5ef0714c8d4edf09d79998fef6674ed4e09/webdev/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/text_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/vision_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/document_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/search_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/webdev/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/text_to_image/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/image_edit/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/text_to_video/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/image_to_video/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/video_edit/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/46919c467f7f93d9609b668a283bcd20d87c28bc/agent/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/ff77a170d5ec578fe4eb8c91e0084a6b1a20beac/text_to_video/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/ff77a170d5ec578fe4eb8c91e0084a6b1a20beac/image_to_video/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/088c607919d1d0d3385409070657f4623e414872/text_to_image/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/088c607919d1d0d3385409070657f4623e414872/image_edit/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/ff77a170d5ec578fe4eb8c91e0084a6b1a20beac/webdev/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/088c607919d1d0d3385409070657f4623e414872/text_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/088c607919d1d0d3385409070657f4623e414872/webdev/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/088c607919d1d0d3385409070657f4623e414872/agent/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/088c607919d1d0d3385409070657f4623e414872/search_style_control/latest-00000-of-00001.parquet",
      "huggingface.co/datasets/lmarena-ai/leaderboard-dataset/blob/088c607919d1d0d3385409070657f4623e414872/vision_style_control/latest-00000-of-00001.parquet"
    ],
    "name": "LMArena leaderboard dataset",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "attribution": "Arena Leaderboard Dataset by LMArena (lmarena-ai), CC BY 4.0. Selected snapshots; ratings rounded to one decimal and transformed into relative scores. No endorsement. Refreshed exact model/task rows from revision 46919c467f7f93d9609b668a283bcd20d87c28bc, observed 2026-10-06. Earlier rows are retained only when their exact entry is absent from the new snapshot. Agent aggregate IPS values from the October 2 snapshot retained in their original signed units; one highest available effort per model, with no component-signal duplication. Latent fit uses a fixed 0.1 IPS unit conversion.",
    "icon": "arena.ai",
    "quote": "Dataset card front matter: license: cc-by-4.0 (not gated)",
    "exact": true
  },
  {
    "id": "artificial-analysis",
    "match": [
      "artificialanalysis.ai/"
    ],
    "name": "Artificial Analysis",
    "kind": "leaderboard",
    "license": "personal, non-commercial use only",
    "terms_url": "https://artificialanalysis.ai/docs/legal/Terms-of-Use.pdf",
    "quote": "§2.1 license to use and access the Site solely for your own personal, noncommercial use; §2.2(c) you shall not access the Site in order to build a similar or competitive website, product, or service",
    "verdict": "not-allowed",
    "checked": "2026-09-25"
  },
  {
    "id": "arena-website",
    "match": [
      "arena.ai/",
      "lmarena.ai/"
    ],
    "name": "LMArena website",
    "kind": "leaderboard",
    "license": "website terms",
    "terms_url": "https://help.arena.ai/articles/5629909088-terms-of-use",
    "quote": "§5: may not use scrapers ... to scrape, extract, or download data ... or commercially exploit the Service or Output. Use the CC BY 4.0 dataset instead.",
    "verdict": "not-allowed",
    "checked": "2026-09-25"
  },
  {
    "id": "epoch-internal",
    "name": "Epoch AI — internal evaluations",
    "match": [
      "epoch.ai/data/benchmark_data.zip"
    ],
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://epoch.ai/benchmarks/use-this-data",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "Epoch AI, Capabilities & Benchmarking (2026), accessed 25 September 2026. Only internal run CSVs; selected settings, rounded results and derived relative scores. External runs excluded.",
    "icon": "epoch.ai",
    "fragments": [
      "chess_puzzles.csv",
      "ebr_bench.csv",
      "frontiermath_erdos.csv",
      "frontiermath_tier_4_v2.csv",
      "frontiermath_tiers_1_3_v2.csv",
      "furniture_assembly.csv",
      "gpqa_diamond.csv",
      "math_level_5.csv",
      "mirrorcode.csv",
      "mystery_game_puzzles.csv",
      "otis_mock_aime_2024_2025.csv",
      "simpleqa_verified.csv",
      "swe_bench_verified.csv"
    ],
    "exact": true
  },
  {
    "id": "open-asr",
    "name": "Open ASR Leaderboard",
    "match": [
      "github.com/huggingface/open_asr_leaderboard/blob/090df1224774bb3c152b57ed3f99898e679222e5/scripts/data/en_longform.csv",
      "github.com/huggingface/open_asr_leaderboard/blob/090df1224774bb3c152b57ed3f99898e679222e5/scripts/data/multilingual.csv",
      "github.com/huggingface/open_asr_leaderboard/blob/16743c6dabc69abcb8fd5bf529a95e3c6c16ac3c/scripts/data/en_shortform.csv"
    ],
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/huggingface/open_asr_leaderboard/blob/main/LICENSE",
    "license_url": "/licenses/open-asr-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "Copyright 2024 The HuggingFace Inc. team. Selected March 2026 CSV suites, WER percentages converted to fractions; model identifiers mapped and scores derived.",
    "icon": "huggingface.co",
    "exact": true
  },
  {
    "id": "models-dev",
    "name": "models.dev — model and provider catalogue",
    "match": [
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3-235b-a22b-instruct-2507.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3-235b-a22b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3-32b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3-coder-next.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3-next-80b-a3b-instruct.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3-next-80b-a3b-thinking.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.5-122b-a10b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.5-27b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.5-35b-a3b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.5-397b-a17b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.5-9b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.6-27b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.6-35b-a3b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.6-plus.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.7-max.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.7-plus.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.8-2.4t-a95b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.8-27b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3.8-flash-next.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/amazon/nova-2-lite.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/amazon/nova-micro.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-fable-5-1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-fable-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-haiku-4-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-opus-4-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-opus-4-6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-opus-4-7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-opus-4-8.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-opus-5-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-opus-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-sonnet-4-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-sonnet-4-6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/anthropic/claude-sonnet-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-r1-distill-qwen-32b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-r1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-v3-0324.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-v3.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-v3.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-v3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-v4-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-v4-pro.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/deepseek/deepseek-v4.1-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-2.5-flash-lite.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-2.5-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-2.5-pro.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-3.1-flash-lite.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-3.5-flash-lite.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-3.5-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-3.6-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-3.7-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemini-3.8-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemma-4-12b-it.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemma-4-E2B-it.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/google/gemma-4-E4B-it.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meituan/longcat-2.0.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/llama-3.1-70b-instruct.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/llama-3.1-8b-instruct.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/llama-3.2-1b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/llama-3.2-3b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/muse-glimmer-30b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/muse-spark-1.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/muse-spark-1.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/meta/muse-spark-1.3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/microsoft/phi-4-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/minimax/MiniMax-M2.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/minimax/MiniMax-M2.7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/minimax/MiniMax-M2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/minimax/MiniMax-M3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/mistral/devstral-small-2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/mistral/magistral-small-2509.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/moonshotai/kimi-k2-thinking.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/moonshotai/kimi-k2.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/moonshotai/kimi-k2.6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/moonshotai/kimi-k3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/llama-3.1-nemotron-ultra-253b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/llama-3.3-nemotron-super-49b-v1.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/nemotron-3-nano-30b-a3b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/nemotron-3-super-120b-a12b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/nemotron-3-ultra-550b-a55b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/nemotron-3.5-lightning.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/nemotron-cascade-2-30b-a3b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/nvidia/nemotron-nano-9b-v2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-4.1-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-4.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5-nano.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.3-codex.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.4-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.4-nano.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.4.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.6-luna.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.6-sol.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.6-terra.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-6-astra.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-6-luna.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-6-sol.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-oss-120b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/gpt-oss-20b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/o3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/o4-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/whisper-large-v3-turbo.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/openai/whisper-large-v3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/stepfun/step-3.5-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/tencent/hy3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/thinkingmachines/inkling-small.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/thinkingmachines/inkling.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/upstage/solar-pro4.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xai/grok-4.3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xai/grok-4.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xai/grok-4.6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xai/grok-4.7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xai/grok-build-0.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xiaomi/mimo-v2-omni.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xiaomi/mimo-v2.5-pro.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/xiaomi/mimo-v2.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-4.5-air.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-4.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-4.6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-4.7-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-4.7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-5.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-5.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-5.3-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-5.3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/zhipuai/glm-image.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/alibaba/models/qwen3-14b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/alibaba/models/qwen3-235b-a22b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/alibaba/models/qwen3-8b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/alibaba/models/qwen3-next-80b-a3b-thinking.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/alibaba/models/qwen3.7-plus.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/apac.amazon.nova-micro-v1:0.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/eu.amazon.nova-2-lite-v1:0.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/meta.llama3-1-8b-instruct-v1:0.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/minimax.minimax-m2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/mistral.magistral-small-2509.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/moonshot.kimi-k2-thinking.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/nvidia.nemotron-super-3-120b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/us.meta.llama3-1-70b-instruct-v1:0.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/amazon-bedrock/models/us.meta.llama3-3-70b-instruct-v1:0.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-fable-5-1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-fable-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-haiku-4-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-opus-4-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-opus-4-6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-opus-4-7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-opus-4-8.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-opus-5-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-opus-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-sonnet-4-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-sonnet-4-6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/anthropic/models/claude-sonnet-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/MiniMaxAI/MiniMax-M2.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/MiniMaxAI/MiniMax-M2.7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/MiniMaxAI/MiniMax-M3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3-32B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3-Next-80B-A3B-Instruct.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.5-122B-A10B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.5-27B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.5-35B-A3B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.5-397B-A17B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.5-9B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.6-35B-A3B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.7-Max.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3.8-2.4T-A95B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/XiaomiMiMo/MiMo-V2.5-Pro.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/XiaomiMiMo/MiMo-V2.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/deepseek-ai/DeepSeek-R1-0528.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/deepseek-ai/DeepSeek-V3-0324.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/deepseek-ai/DeepSeek-V3.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/deepseek-ai/DeepSeek-V3.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/deepseek-ai/DeepSeek-V3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/deepseek-ai/DeepSeek-V4-Flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/deepseek-ai/DeepSeek-V4-Pro.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/moonshotai/Kimi-K2.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/nvidia/Llama-3.3-Nemotron-Super-49B-v1.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/nvidia/Nemotron-3-Nano-30B-A3B.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/tencent/Hy3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/thinkingmachines/Inkling-Small.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/thinkingmachines/Inkling.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepseek/models/deepseek-v4-flash-vision-exp.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/fireworks-ai/models/accounts/fireworks/models/muse-glimmer-30b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/fireworks-ai/models/accounts/fireworks/models/nemotron-lightning-3p5-30b-a3b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-2.5-flash-lite.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-2.5-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-2.5-pro.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-3.1-flash-lite.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-3.1-pro-preview.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-3.5-flash-lite.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-3.5-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-3.6-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-3.7-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/google/models/gemini-3.8-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/groq/models/openai/gpt-oss-120b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/groq/models/openai/gpt-oss-20b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/groq/models/qwen/qwen3.6-27b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/groq/models/qwen/qwen3.8-27b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/meta/models/muse-spark-1.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/meta/models/muse-spark-1.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/moonshotai/models/kimi-k2.6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/moonshotai/models/kimi-k3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-4.1-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-4.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5-nano.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.3-codex.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.4-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.4-nano.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.4.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.6-luna.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.6-sol.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.6-terra.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-6-astra.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-6-luna.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/gpt-6-sol.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/o3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/openai/models/o4-mini.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/togetherai/models/Qwen/Qwen3.6-Plus.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/togetherai/models/deepseek-ai/DeepSeek-R1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/togetherai/models/nvidia/nemotron-3-ultra-550b-a55b.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/upstage/models/solar-pro4.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/xai/models/grok-4.20-0309-reasoning.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/xai/models/grok-4.3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/xai/models/grok-4.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/xai/models/grok-4.6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/xai/models/grok-4.7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-4.5-air.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-4.5.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-4.6.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-4.7-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-4.7.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-5.1.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-5.2.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-5.3-flash.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-5.3.toml",
      "github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/zai/models/glm-5.toml",
      "github.com/anomalyco/models.dev/blob/6b32416340759c52202fef2b0e69cd89adb74bbe/providers/openai/models/gpt-4o-2024-11-20.toml",
      "github.com/anomalyco/models.dev/blob/6b32416340759c52202fef2b0e69cd89adb74bbe/providers/xiaomi/models/mimo-v2.6-pro.toml"
    ],
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/LICENSE",
    "license_url": "/licenses/models-dev-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "Copyright (c) 2025 models.dev. Selected TOML pricing and context fields, identifiers mapped, token prices blended 3:1. OpenRouter ingested data and benchmark fields excluded.",
    "icon": "huggingface.co",
    "exact": true
  },
  {
    "id": "card-bytedance-seed--seed-oss-36b-instruct",
    "name": "ByteDance-Seed/Seed-OSS-36B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/blob/497f1dca95ebdec98e41d517b9f060ee753c902f/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/ByteDance-Seed--Seed-OSS-36B-Instruct.txt",
    "terms_url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/blob/497f1dca95ebdec98e41d517b9f060ee753c902f/LICENSE.txt",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "ByteDance-Seed/Seed-OSS-36B-Instruct authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-minimaxai--minimax-m1-80k",
    "name": "MiniMaxAI/MiniMax-M1-80k model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/MiniMaxAI--MiniMax-M1-80k.txt",
    "terms_url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "huggingface.co",
    "attribution": "MiniMaxAI/MiniMax-M1-80k authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement."
  },
  {
    "id": "card-qwen--qwen3-235b-a22b-instruct-2507",
    "name": "Qwen/Qwen3-235B-A22B-Instruct-2507 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507/blob/ac9c66cc9b46af7306746a9250f23d47083d689e/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-235B-A22B-Instruct-2507.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507/blob/ac9c66cc9b46af7306746a9250f23d47083d689e/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-235B-A22B-Instruct-2507 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the same-family 'Qwen3-235B-A22B Non-thinking' column on the recorded rows, and the TAU1/TAU2 rows for the own model, the Non-thinking column and the '#'-marked (Qwen-generated, GPT-4o-20241120) cells."
  },
  {
    "id": "card-qwen--qwen3-235b-a22b-thinking-2507",
    "name": "Qwen/Qwen3-235B-A22B-Thinking-2507 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507/blob/6cbffae6d8e28b986a6b17bd36f42f9fa0f1f0a5/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-235B-A22B-Thinking-2507.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507/blob/6cbffae6d8e28b986a6b17bd36f42f9fa0f1f0a5/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-235B-A22B-Thinking-2507 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3-30b-a3b-instruct-2507",
    "name": "Qwen/Qwen3-30B-A3B-Instruct-2507 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/0d7cf23991f47feeb3a57ecb4c9cee8ea4a17bfe/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-30B-A3B-Instruct-2507.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/0d7cf23991f47feeb3a57ecb4c9cee8ea4a17bfe/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-30B-A3B-Instruct-2507 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3-30b-a3b-thinking-2507",
    "name": "Qwen/Qwen3-30B-A3B-Thinking-2507 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/blob/144afc2f379b542fdd4e85a1fcd5e1f79112d95d/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-30B-A3B-Thinking-2507.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/blob/144afc2f379b542fdd4e85a1fcd5e1f79112d95d/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-30B-A3B-Thinking-2507 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3-4b-instruct-2507",
    "name": "Qwen/Qwen3-4B-Instruct-2507 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-4B-Instruct-2507/blob/cdbee75f17c01a7cc42f958dc650907174af0554/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-4B-Instruct-2507.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507/blob/cdbee75f17c01a7cc42f958dc650907174af0554/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-4B-Instruct-2507 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement."
  },
  {
    "id": "card-qwen--qwen3-4b-thinking-2507",
    "name": "Qwen/Qwen3-4B-Thinking-2507 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-4B-Thinking-2507/blob/768f209d9ea81521153ed38c47d515654e938aea/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-4B-Thinking-2507.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507/blob/768f209d9ea81521153ed38c47d515654e938aea/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-4B-Thinking-2507 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3-next-80b-a3b-instruct",
    "name": "Qwen/Qwen3-Next-80B-A3B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct/blob/9c7f2fbe84465e40164a94cc16cd30b6999b0cc7/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-Next-80B-A3B-Instruct.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct/blob/9c7f2fbe84465e40164a94cc16cd30b6999b0cc7/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-Next-80B-A3B-Instruct authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3-next-80b-a3b-thinking",
    "name": "Qwen/Qwen3-Next-80B-A3B-Thinking model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking/blob/e502dd4100cc68c0de57643fd4317ec93a128670/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-Next-80B-A3B-Thinking.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking/blob/e502dd4100cc68c0de57643fd4317ec93a128670/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3-Next-80B-A3B-Thinking authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement."
  },
  {
    "id": "card-qwen--qwen3.5-0.8b",
    "name": "Qwen/Qwen3.5-0.8B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-0.8B/blob/2fc06364715b967f1860aea9cf38778875588b17/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-0.8B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-0.8B/blob/2fc06364715b967f1860aea9cf38778875588b17/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-0.8B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3.5-122b-a10b",
    "name": "Qwen/Qwen3.5-122B-A10B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-122B-A10B/blob/dc4d348443bc740c68e2d77492492c11606384d5/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-122B-A10B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B/blob/dc4d348443bc740c68e2d77492492c11606384d5/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-122B-A10B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3.5-27b",
    "name": "Qwen/Qwen3.5-27B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-27B/blob/fc05daec18b0a78c049392ed2e771dde82bdf654/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-27B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-27B/blob/fc05daec18b0a78c049392ed2e771dde82bdf654/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-27B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the Qwen-family columns (Qwen3.5-122B-A10B, Qwen3.5-35B-A3B, Qwen3-235B-A22B = Thinking-2507) on the recorded rows, and the TAU2-Bench and MathVision rows (own, family and footnoted competitor cells)."
  },
  {
    "id": "card-qwen--qwen3.5-2b",
    "name": "Qwen/Qwen3.5-2B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-2B/blob/15852e8c16360a2fea060d615a32b45270f8a8fc/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-2B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-2B/blob/15852e8c16360a2fea060d615a32b45270f8a8fc/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-2B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement."
  },
  {
    "id": "card-qwen--qwen3.5-35b-a3b",
    "name": "Qwen/Qwen3.5-35B-A3B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/59d61f3ce65a6d9863b86d2e96597125219dc754/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-35B-A3B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/59d61f3ce65a6d9863b86d2e96597125219dc754/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-35B-A3B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement."
  },
  {
    "id": "card-qwen--qwen3.5-397b-a17b",
    "name": "Qwen/Qwen3.5-397B-A17B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-397B-A17B/blob/8472618112abcbd45acbcdc58436aff4233c23f7/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-397B-A17B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/blob/8472618112abcbd45acbcdc58436aff4233c23f7/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-397B-A17B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also competitor cells of the TAU2-Bench row and the MathVision row, whose footnotes state all models were evaluated under the card's setup."
  },
  {
    "id": "card-qwen--qwen3.5-4b",
    "name": "Qwen/Qwen3.5-4B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-4B/blob/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-4B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-4B/blob/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-4B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the same-family columns (Qwen3-Next-80B-A3B-Thinking, Qwen3-30B-A3B-Thinking-2507, Qwen3.5-9B) on the recorded rows, and the MathVision row (own, Qwen3.5-9B, GPT-5-Nano, Gemini-2.5-Flash-Lite)."
  },
  {
    "id": "card-qwen--qwen3.5-9b",
    "name": "Qwen/Qwen3.5-9B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-9B/blob/c202236235762e1c871ad0ccb60c8ee5ba337b9a/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.5-9B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-9B/blob/c202236235762e1c871ad0ccb60c8ee5ba337b9a/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.5-9B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3.6-27b",
    "name": "Qwen/Qwen3.6-27B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.6-27B/blob/6a9e13bd6fc8f0983b9b99948120bc37f49c13e9/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.6-27B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.6-27B/blob/6a9e13bd6fc8f0983b9b99948120bc37f49c13e9/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.6-27B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-qwen--qwen3.6-35b-a3b",
    "name": "Qwen/Qwen3.6-35B-A3B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.6-35B-A3B/blob/995ad96eacd98c81ed38be0c5b274b04031597b0/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.6-35B-A3B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B/blob/995ad96eacd98c81ed38be0c5b274b04031597b0/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.6-35B-A3B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: All Qwen-family columns (Qwen3.5-27B, Qwen3.5-35B-A3B, Qwen3.6-35B-A3B) of both tables; Gemma 4 columns only on SWE-bench Pro (refined), NL2Repo, SkillsBench (78-task subset) and QwenClawBench, whose footnotes show Qwen ran every column."
  },
  {
    "id": "card-qwen--qwen3.8-27b",
    "name": "Qwen/Qwen3.8-27B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3.8-27B/blob/1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0/README.md"
    ],
    "exact": true,
    "license": "apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3.8-27B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.8-27B/blob/1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen/Qwen3.8-27B authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-deepseek-ai--deepseek-r1",
    "name": "deepseek-ai/DeepSeek-R1 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/README.md"
    ],
    "exact": true,
    "license": "mit",
    "license_url": "/licenses/deepseek-ai--DeepSeek-R1.txt",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "deepseek-ai/DeepSeek-R1 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Own DeepSeek R1 column and same-lab DeepSeek V3 / R1-Distill cells (unchanged scope; this pass only fills R1 cells the first import skipped)."
  },
  {
    "id": "card-deepseek-ai--deepseek-r1-0528",
    "name": "deepseek-ai/DeepSeek-R1-0528 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-R1-0528/blob/4236a6af538feda4548eca9ab308586007567f52/README.md"
    ],
    "exact": true,
    "license": "mit",
    "license_url": "/licenses/deepseek-ai--DeepSeek-R1-0528.txt",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/blob/4236a6af538feda4548eca9ab308586007567f52/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "deepseek-ai/DeepSeek-R1-0528 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-meituan-longcat--longcat-2.0",
    "name": "meituan-longcat/LongCat-2.0 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/meituan-longcat/LongCat-2.0/blob/834bf5ffe3047aa9f6cc7a64a9bc068b146b8274/README.md"
    ],
    "exact": true,
    "license": "mit",
    "license_url": "/licenses/meituan-longcat--LongCat-2.0.txt",
    "terms_url": "https://huggingface.co/meituan-longcat/LongCat-2.0/blob/834bf5ffe3047aa9f6cc7a64a9bc068b146b8274/LICENSE",
    "quote": "Pinned README declares 'license: mit' and states: 'Any contributions to this repository are licensed under the MIT License, unless otherwise stated.' Evaluation Results: 'Unless noted with `*`, all scores are measured in-house under a unified harness.' Own-model cells and unstarred (in-house) competitor cells are selected; starred cells ('cited from the model's official report') are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "meituan-longcat/LongCat-2.0 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card-microsoft--phi-4-reasoning",
    "name": "microsoft/Phi-4-reasoning model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/microsoft/Phi-4-reasoning/blob/1de18ec97600877ce63dbf60c73b998da99f0195/README.md"
    ],
    "exact": true,
    "license": "mit",
    "license_url": "/licenses/microsoft--Phi-4-reasoning.txt",
    "terms_url": "https://huggingface.co/microsoft/Phi-4-reasoning/blob/1de18ec97600877ce63dbf60c73b998da99f0195/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "microsoft/Phi-4-reasoning authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card-zai-org--glm-5.1",
    "name": "zai-org/GLM-5.1 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/README.md"
    ],
    "exact": true,
    "license": "mit",
    "license_url": "/licenses/zai-org--GLM-5.1.txt",
    "terms_url": "https://huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "zai-org/GLM-5.1 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card-zai-org--glm-5.2",
    "name": "zai-org/GLM-5.2 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/README.md"
    ],
    "exact": true,
    "license": "mit",
    "license_url": "/licenses/zai-org--GLM-5.2.txt",
    "terms_url": "https://huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/LICENSE",
    "quote": "Repository root licence at pinned revision; selected own-model table cells in README only.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "zai-org/GLM-5.2 authors. Selected own-model results, percentages converted to fractions and used to derive relative scores. Full repository licence notice linked; no endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the GLM-5.1 column on HLE, AIME 2026, GPQA-Diamond and SWE-bench Pro, and every column of the MCP-Atlas row (footnote: 'All models were evaluated ...')."
  },
  {
    "id": "unreviewed-hf",
    "name": "Other Hugging Face content",
    "match": [
      "huggingface.co"
    ],
    "kind": "dataset",
    "license": "Repository-specific; not reviewed",
    "terms_url": "https://huggingface.co/terms-of-service",
    "verdict": "unclear",
    "checked": "2026-09-25"
  },
  {
    "id": "gemma4-report",
    "name": "Gemma Team — Gemma 4 Technical Report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2607.02770v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2607.02770v1",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "arxiv.org",
    "attribution": "Gemma Team, Google DeepMind (2026), Gemma 4 Technical Report, arXiv:2607.02770v1, CC BY 4.0. Own-model tables 5 and 6 only; percentages converted to fractions and relative scores derived. Third-party Arena comparison table excluded. No endorsement.",
    "quote": "arXiv HTML licence: CC BY 4.0"
  },
  {
    "id": "amazon-tau2-verified",
    "name": "Amazon AGI — τ²-Bench-Verified",
    "kind": "dataset",
    "match": [
      "github.com/amazon-agi/tau2-bench-verified/blob/864350a8971a8f8ee9e7b8472e2edc380a806b0c/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://github.com/amazon-agi/tau2-bench-verified/blob/864350a8971a8f8ee9e7b8472e2edc380a806b0c/LICENSE",
    "license_url": "/licenses/tau2-verified-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "huggingface.co",
    "attribution": "Amazon AGI, τ²-Bench-Verified repository leaderboard; upstream copyright (c) 2025 Sierra Research. MIT. Selected model configurations, percentages converted to fractions and relative scores derived. Full licence notice linked.",
    "quote": "MIT licence covers software and associated documentation files."
  },
  {
    "id": "massive-agents-paper",
    "name": "MASSIVE-Agents — Kulkarni et al. (2025)",
    "kind": "paper",
    "match": [
      "aclanthology.org/2025.findings-emnlp.1099.pdf"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://aclanthology.org/2025.findings-emnlp.1099/",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "icon": "arxiv.org",
    "attribution": "Mayank Kulkarni, Vittorio Mazzia, Judith Gaspers, Chris Hench, and Jack FitzGerald. MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages. Findings of EMNLP 2025, pp. 20193–20215, ACL. CC BY 4.0. Selected Table 2 AST zero-shot macro averages, percentages converted to fractions; relative scores derived. No endorsement. Table 1 third-party speed data excluded.",
    "quote": "ACL publication footer licenses materials published in or after 2016 under Creative Commons Attribution 4.0."
  },
  {
    "id": "speed-llama-cpp-benches",
    "name": "ggml-org/llama.cpp — committed benchmark documentation",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/ggml-org/llama.cpp/blob/27b20ba8b1b85fc800b0a04caa8f1003ca64a2a3/LICENSE",
    "checked": "2026-09-25",
    "attribution": "The ggml authors. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/ggml-org/llama.cpp/blob/27b20ba8b1b85fc800b0a04caa8f1003ca64a2a3/benches/dgx-spark/dgx-spark.md",
      "github.com/ggml-org/llama.cpp/blob/27b20ba8b1b85fc800b0a04caa8f1003ca64a2a3/benches/mac-m2-ultra/mac-m2-ultra.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices.",
    "license_url": "/licenses/speed-llama-cpp-benches-MIT.txt"
  },
  {
    "id": "speed-onnx-asr",
    "name": "istupakov/onnx-asr — committed benchmark documentation",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/istupakov/onnx-asr/blob/675f0e68c24d846ee1775743e92d9b4ed452380e/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Ilya Stupakov. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/istupakov/onnx-asr/blob/675f0e68c24d846ee1775743e92d9b4ed452380e/docs/benchmarks.md",
      "github.com/istupakov/onnx-asr/blob/675f0e68c24d846ee1775743e92d9b4ed452380e/examples/benchmark-en.ipynb"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices.",
    "license_url": "/licenses/speed-onnx-asr-MIT.txt"
  },
  {
    "id": "speed-apple-coreml",
    "name": "apple/ml-stable-diffusion — committed benchmark documentation",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/apple/ml-stable-diffusion/blob/ea2805dc1945be20561c77e5f6d1d9a5a637cda2/LICENSE.md",
    "checked": "2026-09-25",
    "attribution": "Apple Inc.; benchmarks conducted by Apple and Hugging Face. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/apple/ml-stable-diffusion/blob/ea2805dc1945be20561c77e5f6d1d9a5a637cda2/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices.",
    "license_url": "/licenses/speed-apple-coreml-MIT.txt"
  },
  {
    "id": "speed-magpie-cpp",
    "name": "mudler/magpie-tts.cpp — committed benchmark documentation",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/mudler/magpie-tts.cpp/blob/3008ff73fc2d2da9e4d743b09350aa7023e8980c/LICENSE",
    "checked": "2026-09-25",
    "attribution": "The magpie-tts.cpp authors. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/mudler/magpie-tts.cpp/blob/3008ff73fc2d2da9e4d743b09350aa7023e8980c/benchmarks/BENCHMARK.md",
      "github.com/mudler/magpie-tts.cpp/blob/3008ff73fc2d2da9e4d743b09350aa7023e8980c/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices.",
    "license_url": "/licenses/speed-magpie-cpp-MIT.txt"
  },
  {
    "id": "speed-chatterbox-streaming",
    "name": "davidbrowne17/chatterbox-streaming — committed benchmark documentation",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/davidbrowne17/chatterbox-streaming/blob/41222d1b4f496426fdd6b8cd2e1915fb315e6ff4/LICENSE",
    "checked": "2026-09-25",
    "attribution": "David Browne / chatterbox-streaming contributors. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/davidbrowne17/chatterbox-streaming/blob/41222d1b4f496426fdd6b8cd2e1915fb315e6ff4/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices.",
    "license_url": "/licenses/speed-chatterbox-streaming-MIT.txt"
  },
  {
    "id": "speed-glm-image",
    "name": "zai-org/GLM-Image — committed benchmark documentation",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/zai-org/GLM-Image/blob/69b87db2874f8b556417c03eedf2b8a1484f62e0/LICENSE",
    "checked": "2026-09-25",
    "attribution": "GLM-Image authors, Z.ai. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/zai-org/GLM-Image/blob/69b87db2874f8b556417c03eedf2b8a1484f62e0/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Apache-2.0 sections 1, 2 and 4 cover documentation source, reproduction and derivative works; preserve license/notices and identify modifications.",
    "license_url": "/licenses/speed-glm-image-Apache-2.0.txt"
  },
  {
    "id": "speed-wan22",
    "name": "Wan-Video/Wan2.2 — committed benchmark documentation",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Wan-Video/Wan2.2/blob/1ea34ff48f87168174e12956e200b1d908b1c5ff/LICENSE.txt",
    "checked": "2026-09-25",
    "attribution": "The Alibaba Wan Team Authors. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/Wan-Video/Wan2.2/blob/1ea34ff48f87168174e12956e200b1d908b1c5ff/README.md",
      "github.com/Wan-Video/Wan2.2/blob/1ea34ff48f87168174e12956e200b1d908b1c5ff/assets/comp_effic.png",
      "github.com/Wan-Video/Wan2.2/blob/1ea34ff48f87168174e12956e200b1d908b1c5ff/wan/configs/shared_config.py",
      "github.com/Wan-Video/Wan2.2/blob/1ea34ff48f87168174e12956e200b1d908b1c5ff/wan/configs/wan_t2v_A14B.py",
      "github.com/Wan-Video/Wan2.2/blob/1ea34ff48f87168174e12956e200b1d908b1c5ff/wan/configs/wan_ti2v_5B.py"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Apache-2.0 sections 1, 2 and 4 cover documentation source, reproduction and derivative works; preserve license/notices and identify modifications.",
    "license_url": "/licenses/speed-wan22-Apache-2.0.txt"
  },
  {
    "id": "speed-wan21",
    "name": "Wan-Video/Wan2.1 — committed benchmark documentation",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Wan-Video/Wan2.1/blob/9737cba9c1c3c4d04b33fcad41c111989865d315/LICENSE.txt",
    "checked": "2026-09-25",
    "attribution": "The Alibaba Wan Team Authors. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/Wan-Video/Wan2.1/blob/9737cba9c1c3c4d04b33fcad41c111989865d315/README.md",
      "github.com/Wan-Video/Wan2.1/blob/9737cba9c1c3c4d04b33fcad41c111989865d315/assets/comp_effic.png"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Apache-2.0 sections 1, 2 and 4 cover documentation source, reproduction and derivative works; preserve license/notices and identify modifications.",
    "license_url": "/licenses/speed-wan21-Apache-2.0.txt"
  },
  {
    "id": "quant-llama3-cc0",
    "name": "matt-c1/llama-3-quant-comparison — committed benchmark documentation",
    "kind": "dataset",
    "license": "CC0-1.0",
    "terms_url": "https://github.com/matt-c1/llama-3-quant-comparison/blob/f35b7db21acd1f6258239774f3dbc0fd6cfbfbae/LICENSE",
    "checked": "2026-09-25",
    "attribution": "matt-c1. Selected published measurements, identifiers normalized; reciprocal conversions explicitly marked. No endorsement.",
    "match": [
      "github.com/matt-c1/llama-3-quant-comparison/blob/f35b7db21acd1f6258239774f3dbc0fd6cfbfbae/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Repository CC0 1.0 dedication and fallback license apply to this committed results document; no external database is being relicensed.",
    "license_url": "/licenses/quant-llama3-cc0-CC0-1.0.txt"
  },
  {
    "id": "quant-deepseek-study-v1",
    "name": "Quantitative Analysis of Performance Drop in DeepSeek Model Quantization",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/html/2505.02390v1",
    "checked": "2026-09-25",
    "attribution": "Enbo Zhao et al. (China Unicom), Quantitative Analysis of Performance Drop in DeepSeek Model Quantization, arXiv:2505.02390v1, 5 May 2025, CC BY 4.0. Selected published unweighted Average percentages converted to fractions and ratios. No endorsement.",
    "match": [
      "arxiv.org/html/2505.02390v1#S4.T4"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "Versioned paper itself explicitly declares CC BY 4.0; not inferred from benchmark harness or model licenses.",
    "license_url": "/licenses/deepseek-study-CC-BY-4.0.txt"
  },
  {
    "id": "quant-gemma4-report-v1",
    "name": "Gemma 4 Technical Report v1 — memory and quantization",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2607.02770v1",
    "checked": "2026-09-25",
    "attribution": "Gemma Team, Google DeepMind. Gemma 4 Technical Report, arXiv:2607.02770v1, 2 July 2026, CC BY 4.0. Selected table 3 values transcribed and format labels normalized; raw unit and slash pairs preserved. No endorsement.",
    "match": [
      "arxiv.org/html/2607.02770v1#S2.T3",
      "arxiv.org/html/2607.02770v1#S2.SS5"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned report explicitly declares CC BY 4.0; this is the report grant, not the model-weight Apache license.",
    "license_url": "/licenses/gemma4-CC-BY-4.0.txt"
  },
  {
    "id": "evalplus-published-runs",
    "name": "EvalPlus / EvalPerf published evaluation runs",
    "kind": "dataset",
    "match": [
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/Qwen--Qwen2.5-14B-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/Qwen--Qwen2.5-32B-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/Qwen--Qwen2.5-72B-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/Qwen--Qwen2.5-7B-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/deepseek-chat-v3_openai_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/meta-llama--Llama-3.1-70B-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/meta-llama--Llama-3.1-8B-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/01-ai--Yi-Coder-9B-Chat_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/Qwen--CodeQwen1.5-7B-Chat_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/Qwen--QwQ-32B-Preview_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/claude-3-5-sonnet-20240620_bedrock_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/deepseek-ai--DeepSeek-Coder-V2-Lite-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/deepseek-ai--deepseek-coder-33b-instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/deepseek-ai--deepseek-coder-6.7b-instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/gemini-1.5-flash-002_google_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/gemini-1.5-pro-002_google_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/google--gemma-2-27b-it_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/google--gemma-2-9b-it_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/gpt-4o-2024-08-06_openai_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/gpt-4o-mini-2024-07-18_openai_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/meta-llama--Meta-Llama-3-8B-Instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/microsoft--Phi-3-medium-4k-instruct_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/mistralai--Codestral-22B-v0.1_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/mistralai--Mistral-Large-Instruct-2407_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/mistralai--Mistral-Nemo-Instruct-2407_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/mistralai--Mistral-Small-Instruct-2409_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/nvidia--Llama-3.1-Nemotron-70B-Instruct-HF_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/o1-mini-2024-09-12_openai_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/ise-uiuc--Magicoder-S-DS-6.7B_vllm_temp_1.0_evalperf_results.brief.json",
      "github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/results/evalperf/infly--OpenCoder-8B-Instruct_vllm_temp_1.0_evalperf_results.brief.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/evalplus/evalplus.github.io/blob/87418456d3a8c2bc6265f24ff91d2dd0e7993f4b/LICENSE",
    "license_url": "/licenses/evalplus-published-runs-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "attribution": "EvalPlus authors. Selected published numerical results; model IDs mapped and percentages converted to fractions. No endorsement. Full licence and change notice supplied in independent-NOTICE.txt.",
    "quote": "Root licence in the same pinned repository as these published result files; no nested override found in their paths.",
    "icon": "huggingface.co"
  },
  {
    "id": "aider-polyglot-runs",
    "name": "Aider polyglot benchmark runs",
    "kind": "dataset",
    "match": [
      "github.com/Aider-AI/aider/blob/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/polyglot_leaderboard.yml"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Aider-AI/aider/blob/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/LICENSE.txt",
    "license_url": "/licenses/aider-polyglot-runs-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "Paul Gauthier and Aider contributors. Selected published numerical results; model IDs mapped and percentages converted to fractions. No endorsement. Full licence and change notice supplied in independent-NOTICE.txt.",
    "quote": "Root licence in the same pinned repository as these published result files; no nested override found in their paths.",
    "icon": "huggingface.co"
  },
  {
    "id": "bfcl-result-archive-v4",
    "name": "Berkeley Function Calling Leaderboard result archive",
    "kind": "dataset",
    "match": [
      "github.com/HuanzhiMao/BFCL-Result/blob/cadd76a8dc62984e7bf15ecaf358f070bae745cc/2025-12-16/score/data_overall.csv",
      "github.com/HuanzhiMao/BFCL-Result/blob/cadd76a8dc62984e7bf15ecaf358f070bae745cc/2025-06-14/score/data_overall.csv",
      "github.com/HuanzhiMao/BFCL-Result/blob/cadd76a8dc62984e7bf15ecaf358f070bae745cc/2025-04-25/score/data_overall.csv",
      "github.com/HuanzhiMao/BFCL-Result/blob/cadd76a8dc62984e7bf15ecaf358f070bae745cc/2025-04-13/score/data_overall.csv",
      "github.com/HuanzhiMao/BFCL-Result/blob/cadd76a8dc62984e7bf15ecaf358f070bae745cc/2025-11-03/score/data_overall.csv",
      "github.com/HuanzhiMao/BFCL-Result/blob/cadd76a8dc62984e7bf15ecaf358f070bae745cc/2025-08-26/score/data_overall.csv"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/HuanzhiMao/BFCL-Result/blob/cadd76a8dc62984e7bf15ecaf358f070bae745cc/LICENSE",
    "license_url": "/licenses/bfcl-result-archive-v4-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "attribution": "Huanzhi Mao and the BFCL authors. Selected published numerical results; model IDs mapped and percentages converted to fractions. No endorsement. Full licence and change notice supplied in independent-NOTICE.txt.",
    "quote": "Root licence in the same pinned repository as the imported result files; no nested licence override found. BFCL README additionally explicitly includes evaluation scores.",
    "icon": "huggingface.co"
  },
  {
    "id": "zeroeval-published-results",
    "name": "WildEval ZeroEval published results",
    "kind": "dataset",
    "match": [
      "github.com/WildEval/ZeroEval/blob/8c1485edf12c6efb5f69135a562927c5ad484059/result_dirs/crux.summary.json",
      "github.com/WildEval/ZeroEval/blob/8c1485edf12c6efb5f69135a562927c5ad484059/result_dirs/gsm.summary.json",
      "github.com/WildEval/ZeroEval/blob/8c1485edf12c6efb5f69135a562927c5ad484059/result_dirs/mmlu-redux.summary.json",
      "github.com/WildEval/ZeroEval/blob/8c1485edf12c6efb5f69135a562927c5ad484059/result_dirs/zebra-grid.summary.json",
      "github.com/WildEval/ZeroEval/blob/8c1485edf12c6efb5f69135a562927c5ad484059/result_dirs/math-l5.summary.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/WildEval/ZeroEval/blob/8c1485edf12c6efb5f69135a562927c5ad484059/LICENSE",
    "license_url": "/licenses/zeroeval-published-results-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "attribution": "WildEval / ZeroEval contributors. Selected published numerical results; model IDs mapped and percentages converted to fractions. No endorsement. Full licence and change notice supplied in independent-NOTICE.txt.",
    "quote": "Root licence in the same pinned repository as these published result files; no nested override found in their paths.",
    "icon": "huggingface.co"
  },
  {
    "id": "mmlu-pro-author-runs",
    "name": "MMLU-Pro author-released evaluation runs",
    "kind": "dataset",
    "match": [
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Meta-Llama-3_1-70B-Instruct_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Meta-Llama-3_1-8B-Instruct_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Llama-2-70b-hf_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Meta-Llama-3-8B-Instruct_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Mistral-7B-Instruct-v0.1_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Mistral-7B-Instruct-v0.2_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Mixtral-8x7B-Instruct-v0.1_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Phi-3-mini-4k-instruct_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Qwen1.5-14B-Chat_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Qwen1.5-72B-Chat_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_Qwen1.5-7B-Chat_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_c4ai-command-r-v01_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_claude-3-5-haiku-20241022_5shots.json.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_claude-3-5-sonnet-20241022_5shots.json.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_claude-3.5-sonnet_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_deepseek-chat-v2_5_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_gemini-1.5-flash-002_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_gemini-1.5-pro-002_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_gemini-3.1-pro_5-shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_gpt-4o-2024-08-06_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_gpt-4o-mini_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_gpt4o(2024-05-13)_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_jamba-1.5-large_5shots.zip",
      "github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/eval_results/model_outputs_mathstral-7B_5shots.zip"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/TIGER-AI-Lab/MMLU-Pro/blob/f418b116db00b065c2aea046518d8fcf74d39872/LICENSE",
    "license_url": "/licenses/mmlu-pro-author-runs-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "TIGER-AI-Lab / MMLU-Pro authors. Selected published numerical results; model IDs mapped and percentages converted to fractions. No endorsement. Full licence and change notice supplied in independent-NOTICE.txt.",
    "quote": "Root licence in the same pinned repository as these published result files; no nested override found in their paths.",
    "icon": "huggingface.co"
  },
  {
    "id": "korean-personal-benchmark-runs",
    "name": "Daekeun Kim Korean LLM benchmark runs",
    "kind": "community",
    "match": [
      "github.com/daekeun-ml/evaluate-llm-on-korean-dataset/blob/106ed712b5b88f0e531d5fa503bfed76ac31396e/docs/PREVIOUS_RESULTS.md",
      "github.com/daekeun-ml/evaluate-llm-on-korean-dataset/blob/106ed712b5b88f0e531d5fa503bfed76ac31396e/docs/RESULTS.md"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://github.com/daekeun-ml/evaluate-llm-on-korean-dataset/blob/106ed712b5b88f0e531d5fa503bfed76ac31396e/LICENSE",
    "license_url": "/licenses/korean-personal-benchmark-runs-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2024 Daekeun Kim. Selected published numerical results; model IDs mapped and percentages converted to fractions. No endorsement. Full licence and change notice supplied in independent-NOTICE.txt.",
    "quote": "Root licence in the same pinned repository as these published result files; no nested override found in their paths.",
    "icon": "huggingface.co"
  },
  {
    "id": "litellm-prices",
    "name": "BerriAI / LiteLLM community model price and context catalogue",
    "license": "MIT",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2023 Berri AI. Selected records, identifiers mapped and explicit unit conversions; full MIT notice retained.",
    "kind": "dataset",
    "verdict": "allowed",
    "match": [
      "github.com/BerriAI/litellm/blob/c19ce71bcdad234ca87a7af10cee45375ce811a8/model_prices_and_context_window.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "terms_url": "https://github.com/BerriAI/litellm/blob/c19ce71bcdad234ca87a7af10cee45375ce811a8/LICENSE",
    "quote": "Root LICENSE explicitly licenses content outside enterprise/ under MIT. The root JSON is outside enterprise/; no file-specific exception found in the reviewed file.",
    "license_url": "/licenses/LiteLLM-LICENSE.txt"
  },
  {
    "id": "google-gemini-pricing",
    "name": "Google Gemini Developer API pricing",
    "license": "CC-BY-4.0",
    "checked": "2026-09-25",
    "attribution": "Google, Gemini Developer API pricing, CC BY 4.0, retrieved 25 September 2026. Selected prices and settings, identifiers mapped. No endorsement.",
    "kind": "first-party",
    "verdict": "allowed",
    "match": [
      "ai.google.dev/gemini-api/docs/pricing"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "Page footer explicitly grants CC BY 4.0 to page content except otherwise noted; no selected price-table exception identified. This does not extend to cloud.google.com marketing/pricing pages.",
    "license_url": "/licenses/CC-BY-4.0.txt"
  },
  {
    "id": "cloudflare-workers-ai-pricing",
    "name": "Cloudflare Workers AI pricing",
    "license": "CC-BY-SA-4.0",
    "checked": "2026-09-25",
    "attribution": "Cloudflare and contributors, Workers AI pricing. Selected rates, model identifiers mapped; fixed-setting image costs calculated. Cloudflare-derived extracts and patches are distributed under CC BY-SA 4.0. No endorsement; no trademark licence.",
    "kind": "first-party",
    "verdict": "allowed",
    "match": [
      "github.com/cloudflare/cloudflare-docs/blob/b182511e39518f4b3304edd7bb3e0b0d3017b035/src/content/docs/workers-ai/platform/pricing.mdx"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "terms_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "quote": "Repository README and root LICENSE say CC BY 4.0; repository policy page specifically says Workers documentation CC BY-SA 4.0. Conservative treatment: retain both notices and comply with CC BY-SA 4.0 for this separate tranche. Do not label this MIT or silently remove share-alike.",
    "license_url": "/licenses/CC-BY-SA-4.0.txt"
  },
  {
    "id": "card-deepseek-ai-deepseek-r1",
    "name": "deepseek-ai/DeepSeek-R1 model card",
    "license": "MIT",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/LICENSE",
    "checked": "2026-09-25",
    "attribution": "deepseek-ai, DeepSeek-R1 model card. MIT. Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "Root repository LICENSE covers distributed source documentation, including README.md; source-card declaration agrees. This assessment is for document/data reuse, separate from deploying model weights.",
    "license_url": "/licenses/card-deepseek-ai-deepseek-r1.txt"
  },
  {
    "id": "card-deepseek-ai-deepseek-r1-0528-qwen3-8b",
    "name": "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B model card",
    "license": "MIT",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/LICENSE",
    "checked": "2026-09-25",
    "attribution": "deepseek-ai, DeepSeek-R1-0528-Qwen3-8B model card. MIT. Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "Root repository LICENSE covers distributed source documentation, including README.md; source-card declaration agrees. This assessment is for document/data reuse, separate from deploying model weights.",
    "license_url": "/licenses/card-deepseek-ai-deepseek-r1-0528-qwen3-8b.txt"
  },
  {
    "id": "card-microsoft-phi-4",
    "name": "microsoft/phi-4 model card",
    "license": "MIT",
    "terms_url": "https://huggingface.co/microsoft/phi-4/blob/2db69c1c3e91a05d2c64a3185acfbaf36f744e25/LICENSE",
    "checked": "2026-09-25",
    "attribution": "microsoft, phi-4 model card. MIT. Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/microsoft/phi-4/blob/2db69c1c3e91a05d2c64a3185acfbaf36f744e25/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "Root repository LICENSE covers distributed source documentation, including README.md; source-card declaration agrees. This assessment is for document/data reuse, separate from deploying model weights.",
    "license_url": "/licenses/card-microsoft-phi-4.txt"
  },
  {
    "id": "card-microsoft-phi-4-mini-instruct",
    "name": "microsoft/Phi-4-mini-instruct model card",
    "license": "MIT",
    "terms_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/LICENSE",
    "checked": "2026-10-09",
    "attribution": "microsoft, Phi-4-mini-instruct model card. MIT. Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "Root repository LICENSE covers distributed source documentation, including README.md; source-card declaration agrees. This assessment is for document/data reuse, separate from deploying model weights.",
    "license_url": "/licenses/card-microsoft-phi-4-mini-instruct.txt",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-microsoft-phi-4-mini-reasoning",
    "name": "microsoft/Phi-4-mini-reasoning model card",
    "license": "MIT",
    "terms_url": "https://huggingface.co/microsoft/Phi-4-mini-reasoning/blob/0e3b1e2d02ee478a3743abe3f629e9c0cb722e0a/LICENSE",
    "checked": "2026-10-09",
    "attribution": "microsoft, Phi-4-mini-reasoning model card. MIT. Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/microsoft/Phi-4-mini-reasoning/blob/0e3b1e2d02ee478a3743abe3f629e9c0cb722e0a/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "Root repository LICENSE covers distributed source documentation, including README.md; source-card declaration agrees. This assessment is for document/data reuse, separate from deploying model weights.",
    "license_url": "/licenses/card-microsoft-phi-4-mini-reasoning.txt",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card-nvidia-nemotron-cascade-2-30b-a3b",
    "name": "nvidia/Nemotron-Cascade-2-30B-A3B model card",
    "license": "NVIDIA Open Model License (2025-10-24)",
    "terms_url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/",
    "checked": "2026-09-25",
    "attribution": "nvidia, Nemotron-Cascade-2-30B-A3B model card. NVIDIA Open Model License (2025-10-24). Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B/blob/6327cdbcf907e1c7cec9cb29fb6e6cebdf8feaf7/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "The Work/Model definition explicitly includes documentation. The licence permits commercial reproduction, display, derivatives and redistribution subject to its conditions; full terms and required notice retained. No model weights are redistributed.",
    "license_url": "/licenses/card-nvidia-nemotron-cascade-2-30b-a3b.txt"
  },
  {
    "id": "card-nvidia-nvidia-nemotron-3-nano-30b-a3b-bf16",
    "name": "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 model card",
    "license": "NVIDIA Nemotron Open Model License (2025-12-15)",
    "terms_url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/",
    "checked": "2026-09-25",
    "attribution": "nvidia, NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 model card. NVIDIA Nemotron Open Model License (2025-12-15). Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "The Work/Model definition explicitly includes documentation. The licence permits commercial reproduction, display, derivatives and redistribution subject to its conditions; full terms and required notice retained. No model weights are redistributed.",
    "license_url": "/licenses/card-nvidia-nvidia-nemotron-3-nano-30b-a3b-bf16.txt"
  },
  {
    "id": "card-nvidia-nvidia-nemotron-3-nano-4b-bf16",
    "name": "nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 model card",
    "license": "NVIDIA Nemotron Open Model License (2025-12-15)",
    "terms_url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/",
    "checked": "2026-09-25",
    "attribution": "nvidia, NVIDIA-Nemotron-3-Nano-4B-BF16 model card. NVIDIA Nemotron Open Model License (2025-12-15). Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16/blob/dfaf35de3e30f1867dd8dbc38a7fc9fb52d3914f/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "The Work/Model definition explicitly includes documentation. The licence permits commercial reproduction, display, derivatives and redistribution subject to its conditions; full terms and required notice retained. No model weights are redistributed.",
    "license_url": "/licenses/card-nvidia-nvidia-nemotron-3-nano-4b-bf16.txt"
  },
  {
    "id": "card-nvidia-nvidia-nemotron-3-super-120b-a12b-bf16",
    "name": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 model card",
    "license": "NVIDIA Nemotron Open Model License (2025-12-15)",
    "terms_url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/",
    "checked": "2026-09-25",
    "attribution": "nvidia, NVIDIA-Nemotron-3-Super-120B-A12B-BF16 model card. NVIDIA Nemotron Open Model License (2025-12-15). Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/blob/2dc98e2afe4face0e4ce40972a915c45368bd34a/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "The Work/Model definition explicitly includes documentation. The licence permits commercial reproduction, display, derivatives and redistribution subject to its conditions; full terms and required notice retained. No model weights are redistributed.",
    "license_url": "/licenses/card-nvidia-nvidia-nemotron-3-super-120b-a12b-bf16.txt"
  },
  {
    "id": "card-nvidia-nvidia-nemotron-nano-9b-v2",
    "name": "nvidia/NVIDIA-Nemotron-Nano-9B-v2 model card",
    "license": "NVIDIA Open Model License (2025-10-24)",
    "terms_url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/",
    "checked": "2026-09-25",
    "attribution": "nvidia, NVIDIA-Nemotron-Nano-9B-v2 model card. NVIDIA Open Model License (2025-10-24). Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2/blob/6533e8de2c68e4536bf7c411d7a3ce5734111476/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "The Work/Model definition explicitly includes documentation. The licence permits commercial reproduction, display, derivatives and redistribution subject to its conditions; full terms and required notice retained. No model weights are redistributed.",
    "license_url": "/licenses/card-nvidia-nvidia-nemotron-nano-9b-v2.txt"
  },
  {
    "id": "card-qwen-qwen3-coder-30b-a3b-instruct",
    "name": "Qwen/Qwen3-Coder-30B-A3B-Instruct model card",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/b2cff646eb4bb1d68355c01b18ae02e7cf42d120/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen, Qwen3-Coder-30B-A3B-Instruct model card. Apache-2.0. Selected benchmark results; changed to normalized numerical JSON. No endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/b2cff646eb4bb1d68355c01b18ae02e7cf42d120/README.md"
    ],
    "icon": "huggingface.co",
    "quote": "Root repository LICENSE covers distributed source documentation, including README.md; source-card declaration agrees. This assessment is for document/data reuse, separate from deploying model weights.",
    "license_url": "/licenses/card-qwen-qwen3-coder-30b-a3b-instruct.txt"
  },
  {
    "id": "report-2512.13961v2",
    "name": "Olmo 3",
    "license": "CC BY 4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "checked": "2026-09-25",
    "attribution": "Team Olmo / Allen Institute for AI. Olmo 3. arXiv:2512.13961v2. CC BY 4.0. Selected numerical data, percent-to-fraction conversion and labels normalized; no endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "arxiv.org/html/2512.13961v2"
    ],
    "icon": "huggingface.co",
    "quote": "The pinned report itself explicitly declares CC BY 4.0. This licence applies to report contents, regardless of a different model-weight licence.",
    "license_url": "/licenses/report-2512.13961v2.txt"
  },
  {
    "id": "report-2601.02346v1",
    "name": "Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling",
    "license": "CC BY 4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "checked": "2026-10-09",
    "attribution": "Falcon-H1R authors / Technology Innovation Institute. Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling. arXiv:2601.02346v1. CC BY 4.0. Selected numerical data, percent-to-fraction conversion and labels normalized; no endorsement.",
    "verdict": "allowed",
    "kind": "first-party",
    "exact": true,
    "match": [
      "arxiv.org/html/2601.02346v1"
    ],
    "icon": "huggingface.co",
    "quote": "The pinned report itself explicitly declares CC BY 4.0. This licence applies to report contents, regardless of a different model-weight licence.",
    "license_url": "/licenses/report-2601.02346v1.txt",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "speech-ttsds-v1-mit",
    "name": "TTSDS v1 published results",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://huggingface.co/datasets/ttsds/results/blob/111ccecaa50aec4ca1247c5f5d95c2058f6f2946/README.md",
    "checked": "2026-09-25",
    "attribution": "TTSDS / Christoph Minixhofer and contributors. MIT. Preserve the MIT notice; this extraction derives the default equal-factor composite from published result rows.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/datasets/ttsds/results/blob/111ccecaa50aec4ca1247c5f5d95c2058f6f2946/results.csv"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Dataset card explicitly licenses this results dataset MIT; aggregation implementation also MIT.",
    "license_url": "/licenses/speech-ttsds-v1-mit.txt"
  },
  {
    "id": "speech-emergent-apache",
    "name": "EmergentTTS-Eval six-category leaderboard",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/boson-ai/EmergentTTS-Eval-public/blob/fe02c5d27f2d87751a66e32631aaa73a92d9cbbb/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Boson AI and EmergentTTS-Eval authors. Apache License 2.0. Selected README result rows, percentages divided by 100; retain LICENSE.",
    "verdict": "allowed",
    "match": [
      "github.com/boson-ai/EmergentTTS-Eval-public/blob/fe02c5d27f2d87751a66e32631aaa73a92d9cbbb/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Repository README numerical leaderboard under root Apache-2.0 grant; no conflicting result-file licence found.",
    "license_url": "/licenses/speech-emergent-apache.txt"
  },
  {
    "id": "speech-voxtral-tts-report",
    "name": "Voxtral TTS",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2603.25551v2",
    "checked": "2026-09-25",
    "attribution": "Mistral AI et al.; Voxtral TTS; arXiv:2603.25551v2. CC BY 4.0. Numerical tables transcribed; percentage metrics divided by 100; explicitly marked macro means derived. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2603.25551v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned report, including numerical tables. Not a claim about model-weight or underlying audio licences.",
    "license_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "speech-fish-s2-report",
    "name": "Fish Audio S2 Technical Report",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2603.08823v2",
    "checked": "2026-09-25",
    "attribution": "Fish Audio Team (Shijia Liao et al.); Fish Audio S2 Technical Report; arXiv:2603.08823v2. CC BY 4.0. Numerical tables transcribed; percentage metrics divided by 100; explicitly marked macro means derived. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2603.08823v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned report, including numerical tables. Not a claim about model-weight or underlying audio licences.",
    "license_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "speech-dots-tts-report",
    "name": "dots.tts Technical Report",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2606.07080v2",
    "checked": "2026-10-09",
    "attribution": "dots.tts Team (Shi Lian et al.); dots.tts Technical Report; arXiv:2606.07080v2. CC BY 4.0. Numerical tables transcribed; percentage metrics divided by 100; explicitly marked macro means derived. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2606.07080v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned report, including numerical tables. Not a claim about model-weight or underlying audio licences.",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "scope": " Extended 2026-10-09: also the dots.tts (Pretrain) row (WER/CER/SIM, test-en and test-zh; mapped to dots-tts-base) and the SOAR row's SIM cells; other makers' rows not used. Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-E-2026-10-09.md."
  },
  {
    "id": "speech-voxtral-realtime-report",
    "name": "Voxtral Realtime",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2602.11298v1",
    "checked": "2026-09-25",
    "attribution": "Mistral AI et al.; Voxtral Realtime; arXiv:2602.11298v1. CC BY 4.0. Numerical tables transcribed; percentage metrics divided by 100; explicitly marked macro means derived. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2602.11298v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned report, including numerical tables. Not a claim about model-weight or underlying audio licences.",
    "license_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "speech-gigaspeechbench-report",
    "name": "GigaSpeechBench: A Real-World Multilingual Speech-to-Text Benchmark",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2606.28884v1",
    "checked": "2026-09-25",
    "attribution": "Yujie Tu et al.; GigaSpeechBench: A Real-World Multilingual Speech-to-Text Benchmark; arXiv:2606.28884v1. CC BY 4.0. Numerical tables transcribed; percentage metrics divided by 100; explicitly marked macro means derived. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2606.28884v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned report, including numerical tables. Not a claim about model-weight or underlying audio licences. The project repository/dataset lacked an explicit grant at review, so only this CC BY report supplies the numbers.",
    "license_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "speech-stt-ru-mit",
    "name": "Russian STT benchmark by domain",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/vakovalskii/stt-ru-benchmark/blob/387548b8d9fdf51791a837b0c42059d2c2ebcaa9/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2026 Valerii Kovalskii. MIT. Retain LICENSE; WER verified against committed transcripts with documented basic normalization.",
    "verdict": "allowed",
    "match": [
      "github.com/vakovalskii/stt-ru-benchmark/blob/387548b8d9fdf51791a837b0c42059d2c2ebcaa9/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Root MIT licence applies to repository result files and README; numerical data reused, not source audio redistributed.",
    "license_url": "/licenses/speech-stt-ru-mit.txt"
  },
  {
    "id": "speech-smolevich-memo-licensed",
    "name": "Live voice-memo STT pilot",
    "kind": "dataset",
    "license": "MIT; README findings CC BY 4.0",
    "terms_url": "https://github.com/Smolevich/stt-benchmarks/blob/b6349a817a84cb0d74369b02ea951f1d6112f4f2/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2026 Stanislav Shupilkin / Stas Shupilkin. Repository MIT; README findings and methodology CC BY 4.0. Preserve notices and credit Smolevich/stt-benchmarks. Full-live-set medians selected and divided by 100.",
    "verdict": "allowed",
    "match": [
      "github.com/Smolevich/stt-benchmarks/blob/b6349a817a84cb0d74369b02ea951f1d6112f4f2/results/summary.csv"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Root MIT repository grant; README separately explicitly grants CC BY 4.0 for findings/methodology. No reuse or redistribution of private audio.",
    "license_url": "/licenses/speech-smolevich-memo-licensed.txt"
  },
  {
    "id": "speech-cn-news-tts-ccby",
    "name": "CN-NewsTTS Bench v0.1",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://github.com/Jayden-X-L/cn-news-tts-bench/blob/5cd2dcf0896f2391b877317221892a24066588bb/LICENSE-DATA.md",
    "checked": "2026-09-25",
    "attribution": "CN-NewsTTS Bench by Jayden-X-L. CC BY 4.0. Selected strict target-accuracy results, original fractional scale retained. Credit/link repository; no endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/Jayden-X-L/cn-news-tts-bench/blob/5cd2dcf0896f2391b877317221892a24066588bb/artifacts/scores/leaderboard.csv"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "LICENSE-DATA.md explicitly covers benchmark data, fixed ASR transcripts, public leaderboard results and documentation. Code separately MIT. No source audio reused.",
    "license_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "visual-mage-flow",
    "name": "Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and Editing",
    "match": [
      "arxiv.org/html/2607.19064v2"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-mage-flow.txt",
    "attribution": "Xinjie Zhang , Peng Zhang , Shicheng Zheng , Jinghao Guo , Zhaoyang Jia , Yifei Shen , Xun Guo , Yuxuan Luo , Jiahao Li , Wenxuan Xie , Fanyi Pu , Xiaoyi Zhang , Kaichen Zhang , Zongyu Guo , Tianci Bi , Dongnan Gui , Zhening Liu , Zimo Wen , Zihan Zheng , Senqiao Yang , Xiao Li , Jinglu Wang , Bin Li , Yan Lu. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://arxiv.org/html/2607.19064v2. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Only the three authors’ Mage-Flow checkpoints. No baseline measurement is promoted.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-humanscore",
    "name": "HumanScore: Benchmarking Human Motions in Generated Videos",
    "match": [
      "arxiv.org/html/2604.20157v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-humanscore.txt",
    "attribution": "Yusu Fang , Tiange Xiang , Tian Tan , Narayan Schuetz , Scott Delp , Li Fei-Fei , Ehsan Adeli. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://arxiv.org/html/2604.20157v1. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Original motion benchmark evaluations; not external VBench/Arena/AA values.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-physics-verified",
    "name": "Physics-IQ Verified",
    "match": [
      "arxiv.org/html/2606.18943v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-physics-verified.txt",
    "attribution": "Tim Rädsch , Yuki M Asano , Hilde Kuehne , Stefan Bauer , Priyank Jaini , Robert Geirhos , Carsten T. Lüth. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://arxiv.org/html/2606.18943v1. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Original factorial experiment; choose verified GT + verified metric + bpp only. Affiliations include Google DeepMind; evaluated models are other providers.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-vgi",
    "name": "VGI-Bench: Probing Visual Intelligence in Video Generation Models",
    "match": [
      "arxiv.org/html/2608.19583v3"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-vgi.txt",
    "attribution": "Xuan He , Cong Wei , Yuhao Cheng , Linrui Ma , Yuxuan Zhang , Zuojun Li , Yuhao Wen , Jize Jiang , Zeyi Liu , Yuren Hao , Songcheng Cai , Keming Wu , Penghui Du , Kai Zou , Rui Yang , Chenkai Sun , Ke Yang , Ping Nie , Kelsey R Allen , Chenglong Wang , Michel Galley , Jianfeng Gao , ChengXiang Zhai. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://arxiv.org/html/2608.19583v3. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Original I2V evaluation, not full-instance coverage. Appendix B.1 resolves Wan2.2-I2V-A14B. MiniMax-H3 row included (sweep-visual 2026-09-29): in the pinned v3 HTML, Table 2's 'Open Source' header (colspan 3) covers Mnx-H3, HY1.5 and Wan2.2, matching Table 10's Open Source block (960×544, no price), so the commercial/open label conflict noted earlier is not present in v3.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-rapid-physics",
    "name": "Rapidata/world-model-physics published dataset-card rankings",
    "match": [
      "huggingface.co/datasets/Rapidata/world-model-physics/blob/cf0e0724866405b3b7fb8faae3e59b9f7569138b/README.md"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-rapid-physics.txt",
    "attribution": "Rapidata. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://huggingface.co/datasets/Rapidata/world-model-physics/blob/cf0e0724866405b3b7fb8faae3e59b9f7569138b/README.md. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Rapidata’s own collected human comparisons and published with-reference Elo in the CC-BY dataset card. No external live website extraction.",
    "icon": "huggingface.co"
  },
  {
    "id": "visual-rapid-camera",
    "name": "Rapidata/camera-movement published dataset-card rankings",
    "match": [
      "huggingface.co/datasets/Rapidata/camera-movement/blob/7e12622bf6a39b07bfead20e2cede1ae056459ea/README.md",
      "huggingface.co/datasets/Rapidata/camera-movement-human-preference-324k/blob/6206acfab997d39eaf63012d127c917de7995cc9/README.md"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "license_url": "/licenses/visual-rapid-camera.txt",
    "attribution": "Rapidata. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://huggingface.co/datasets/Rapidata/camera-movement/blob/7e12622bf6a39b07bfead20e2cede1ae056459ea/README.md. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Rapidata’s own human annotations and published Movement Elo. Dataset card explicitly licenses human annotations CC-BY-4.0; generated clips have separate terms and are not redistributed.",
    "icon": "huggingface.co"
  },
  {
    "id": "visual-geneval",
    "name": "GenEval authors' results: djghosh13/geneval README and arXiv 2310.11513v1",
    "match": [
      "github.com/djghosh13/geneval/blob/af4902f24d3ca90ebbb446dd9891a59e0f82725f/README.md",
      "arxiv.org/html/2310.11513v1"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "MIT (README); CC-BY-4.0 (arXiv 2310.11513v1)",
    "terms_url": "https://github.com/djghosh13/geneval/blob/af4902f24d3ca90ebbb446dd9891a59e0f82725f/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-geneval.txt",
    "attribution": "djghosh13/geneval contributors (README, MIT) and Dhruba Ghosh, Hanna Hajishirzi, Ludwig Schmidt, GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment, arXiv:2310.11513v1 (17 Oct 2023), CC BY 4.0. Extracted selected measured table values; model labels mapped only to the stated model version. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "README: MIT repository licence (LICENSE at commit af4902f). Paper: arXiv abs page for 2310.11513v1 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2310.11513v1\". Both publish the same GenEval-author experiment; README rows are not duplicated from the paper.",
    "scope": "Original benchmark creators' own evaluation: README table and the paper's Appendix Table 3 (all SD v1.x/v2.x versions, SD-XL 1.0 and IF-M/L/XL they ran). minDALL-E withheld (min-dalle port; Mini vs Mega weights not stated). CLIP retrieval baseline is not a model.",
    "icon": "huggingface.co"
  },
  {
    "id": "visual-hidream",
    "name": "HiDream-ai/HiDream-O1-Image repository README results",
    "match": [
      "github.com/HiDream-ai/HiDream-O1-Image/blob/2c2d29ff729e48f33e41f49edfdbd81d5ac103b4/README.md"
    ],
    "exact": true,
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://github.com/HiDream-ai/HiDream-O1-Image/blob/2c2d29ff729e48f33e41f49edfdbd81d5ac103b4/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "license_url": "/licenses/visual-hidream.txt",
    "attribution": "HiDream-ai/HiDream-O1-Image contributors. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://github.com/HiDream-ai/HiDream-O1-Image/blob/2c2d29ff729e48f33e41f49edfdbd81d5ac103b4/README.md. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Own-model rows only. MIT includes associated documentation in this repository; no external leaderboard database is used. Extended 2026-10-09: HiDream's own models only (HiDream-O1-Image, plus HiDream-I1-Full in the DPG-Bench table); other makers' rows not used. Also the LongText-Bench table's two HiDream rows (HiDream-O1-Image, HiDream-I1-Full). Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-E-2026-10-09.md.",
    "icon": "huggingface.co"
  },
  {
    "id": "visual-glm-image",
    "name": "zai-org/GLM-Image repository README results",
    "match": [
      "github.com/zai-org/GLM-Image/blob/69b87db2874f8b556417c03eedf2b8a1484f62e0/README.md"
    ],
    "exact": true,
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/zai-org/GLM-Image/blob/69b87db2874f8b556417c03eedf2b8a1484f62e0/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-glm-image.txt",
    "attribution": "zai-org/GLM-Image contributors. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://github.com/zai-org/GLM-Image/blob/69b87db2874f8b556417c03eedf2b8a1484f62e0/README.md. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Own-model table cells in the Apache-2.0 repository README. Baselines and unclear TIIF split excluded.",
    "icon": "huggingface.co"
  },
  {
    "id": "visual-ernie-image",
    "name": "ERNIE-Image model repository evaluation tables",
    "match": [
      "huggingface.co/baidu/ERNIE-Image/blob/5346b31d68c9c23758ba56ef8be5e9dc174c7f99/README.md"
    ],
    "exact": true,
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/baidu/ERNIE-Image/blob/5346b31d68c9c23758ba56ef8be5e9dc174c7f99/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-ernie-image.txt",
    "attribution": "Baidu. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://huggingface.co/baidu/ERNIE-Image/blob/5346b31d68c9c23758ba56ef8be5e9dc174c7f99/README.md. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Own-model rows only, explicitly without prompt enhancement. Apache-2.0 model repository card, not a third-party website.",
    "icon": "huggingface.co"
  },
  {
    "id": "visual-qwenbench",
    "name": "Qwen-Image-Bench: From Generation to Creation in Text-to-Image Evaluation",
    "match": [
      "arxiv.org/html/2605.28091v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-qwenbench.txt",
    "attribution": "Niantong Li , Guangzheng Hu , Weixu Qiao , Ying Ba , Qichen Hong , Shijun Shen , Jinlin Wang , Fan Zhou , Jianye Kang , Xin Shang , Ziyi He , Wei Wang , Dalin Li , Jiahao Li , Jie Zhang , Kaiyuan Gao , Kun Yan , Lihan Jiang , Ningyuan Tang , Shengming Yin , Tianhe Wu , Xiao Xu , Xiaoyue Chen , Yuxiang Chen , Yan Shu , Yanran Zhang , Yilei Chen , Yixian Xu , Zekai Zhang , Zhendong Wang , Zihao Liu , Zikai Zhou , Hongzhu Shi , Yi Wang , Bing Zhao , Hu Wei , Lin Qu , Chenfei Wu. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://arxiv.org/html/2605.28091v1. No endorsement; provided as-is.",
    "quote": "Only the retained original measurements and accompanying documentation inside this licensed artifact. No permission inferred for copied third-party comparison rows, external websites, model weights, generated media or dataset inputs.",
    "scope": "Explicit same-evaluator original runs. Qwen models lab; other providers independent of model author, although the evaluator is a competing provider.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-llada-image",
    "name": "LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes",
    "match": [
      "arxiv.org/html/2609.03796v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-SA-4.0",
    "terms_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "license_url": "/licenses/visual-llada-image.txt",
    "attribution": "Chuyan Chen , Haoxing Chen , Kun Chen , Zhenglin Cheng , Long Cui , Ruishan Fang , Zhangxuan Gu , Zhicheng Huang , Zhenzhong Lan , Yuanting Lei , Haoquan Li , Jianguo Li , Rongchuan Li , Sidu Li , Tao Lin , Deyuan Liu , Jiacheng Liu , Lin Liu , Yuxuan Lou , Zhisheng Lu , Yuxin Ma , Shuheng Shen , Peng Sun , Chaoyang Wang , Hongjun Wang , Xiaomei Wang , Yongxin Wang , Chengzhang Wu , Hongru Wu , Jun Xie. Extracted selected measured table values; percentages divided by 100 where specified; model labels mapped only to the stated model version. No values estimated or rerun. Source: https://arxiv.org/html/2609.03796v1. No endorsement; provided as-is.",
    "quote": "Own-model measurements extracted and adapted from arXiv:2609.03796v1, redistributed in the separately licensed visual-cc-by-sa.json tranche.",
    "scope": "Only the LLaDA authors’ own LLaDA-Image and LLaDA-Image Turbo rows in Tables 7 and 8. Qwen comparison rows excluded.",
    "icon": "arxiv.org"
  },
  {
    "id": "livecodebench-website-20241226",
    "name": "LiveCodeBench official website results (2024-12-26)",
    "kind": "dataset",
    "match": [
      "github.com/LiveCodeBench/livecodebench.github.io/blob/aa37a9f8317713b146ff5ada8ab8b2513bb20b70/src/mocks/performances_generation.json"
    ],
    "exact": true,
    "license": "CC-BY-SA-4.0",
    "license_url": "/licenses/lcb-sharealike-NOTICE.txt",
    "terms_url": "https://github.com/LiveCodeBench/livecodebench.github.io/blob/aa37a9f8317713b146ff5ada8ab8b2513bb20b70/README.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "LiveCodeBench website evaluation results by the LiveCodeBench authors (Naman Jain, King Han, Alex Gu, Wen-Ding Li, Fanjia Yan, Tianjun Zhang, Sida Wang, Armando Solar-Lezama, Koushik Sen, Ion Stoica). CC BY-SA 4.0. Adapted by whatmodel.sh: selected models and explicit date windows, reproduced the website arithmetic mean and one-decimal percentage rounding, mapped catalogue IDs, converted percentages to fractions, and added provenance. No endorsement.",
    "quote": "This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.",
    "adapter_license": "CC-BY-SA-4.0",
    "adapter_license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "notice_file": "/licenses/lcb-sharealike-NOTICE.txt",
    "adapted_data_file": "/data/lcb-sharealike.json",
    "provenance_file": "/data/lcb-sharealike-provenance.json",
    "disclaimer": "Licensed material is provided as-is and as-available, without warranties; see CC BY-SA 4.0 section 5 for the complete disclaimer and liability limitation.",
    "license_scope": "Numerical website release files in this same repository; not inferred from the MIT harness.",
    "publication_requirement": "Retain this tranche, attribution, changes, licence and provenance when displaying or serving its rows.",
    "icon": "huggingface.co"
  },
  {
    "id": "livecodebench-website-20250801",
    "name": "LiveCodeBench official website results (2025-08-01)",
    "kind": "dataset",
    "match": [
      "github.com/LiveCodeBench/livecodebench.github.io/blob/f186ab5041d6b768733b8ae8145151c867942134/build/performances_generation.json",
      "github.com/LiveCodeBench/livecodebench.github.io/blob/f186ab5041d6b768733b8ae8145151c867942134/build/v5.json"
    ],
    "exact": true,
    "license": "CC-BY-SA-4.0",
    "license_url": "/licenses/lcb-sharealike-NOTICE.txt",
    "terms_url": "https://github.com/LiveCodeBench/livecodebench.github.io/blob/f186ab5041d6b768733b8ae8145151c867942134/README.md",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "attribution": "LiveCodeBench website evaluation results by the LiveCodeBench authors (Naman Jain, King Han, Alex Gu, Wen-Ding Li, Fanjia Yan, Tianjun Zhang, Sida Wang, Armando Solar-Lezama, Koushik Sen, Ion Stoica). CC BY-SA 4.0. Adapted by whatmodel.sh: selected models and explicit date windows, reproduced the website arithmetic mean and one-decimal percentage rounding, mapped catalogue IDs, converted percentages to fractions, and added provenance. No endorsement.",
    "quote": "This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.",
    "adapter_license": "CC-BY-SA-4.0",
    "adapter_license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "notice_file": "/licenses/lcb-sharealike-NOTICE.txt",
    "adapted_data_file": "/data/lcb-sharealike.json",
    "provenance_file": "/data/lcb-sharealike-provenance.json",
    "disclaimer": "Licensed material is provided as-is and as-available, without warranties; see CC BY-SA 4.0 section 5 for the complete disclaimer and liability limitation.",
    "license_scope": "Numerical website release files in this same repository; not inferred from the MIT harness.",
    "publication_requirement": "Retain this tranche, attribution, changes, licence and provenance when displaying or serving its rows.",
    "icon": "huggingface.co"
  },
  {
    "id": "card-ifm-k2-horizon-7b",
    "name": "IFM/K2-Horizon-7B model card",
    "match": [
      "huggingface.co/IFM/K2-Horizon-7B/blob/036114ce8d46c32b24c15423211069abb9c5d25e/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "IFM/K2-Horizon-7B — IFM/K2-Horizon-7B model card\nSource: https://huggingface.co/IFM/K2-Horizon-7B/blob/036114ce8d46c32b24c15423211069abb9c5d25e/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to IFM. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-liquidai-lfm2-5-1-2b-thinking",
    "name": "LiquidAI/LFM2.5-1.2B-Thinking model card",
    "match": [
      "huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking/blob/f313478934a7612d22991f752959d7a1a8756fec/README.md"
    ],
    "license": "LFM Open License v1.0",
    "terms_url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking/blob/f313478934a7612d22991f752959d7a1a8756fec/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "Liquid AI — LFM2.5-1.2B-Thinking model card. Source: https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking/blob/f313478934a7612d22991f752959d7a1a8756fec/README.md\nLicensed under LFM Open License v1.0: https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking/blob/f313478934a7612d22991f752959d7a1a8756fec/LICENSE\nModified for whatmodel.sh: selected benchmark cells, normalized percentages to fractions, and clarified metric/setup labels. No endorsement. Commercial reuse remains subject to the licence’s annual-revenue condition and other terms.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "The explicitly linked LFM Open License v1.0 includes documentation and grants reproduction, adaptation, display and distribution subject to all its terms. Section 5 imposes commercial revenue and Legal Entity conditions; each recipient must assess eligibility.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice.",
    "conditions": "LFM Open License v1.0 includes commercial revenue and Legal Entity conditions. Downstream users must assess their own eligibility; this compilation does not remove these restrictions."
  },
  {
    "id": "card-liquidai-lfm2-5-8b-a1b",
    "name": "LiquidAI/LFM2.5-8B-A1B model card",
    "match": [
      "huggingface.co/LiquidAI/LFM2.5-8B-A1B/blob/5dd22602c2e9f6a097b1de4c4efe0658b605015c/README.md"
    ],
    "license": "LFM Open License v1.0",
    "terms_url": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B/blob/5dd22602c2e9f6a097b1de4c4efe0658b605015c/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "Liquid AI — LFM2.5-8B-A1B model card. Source: https://huggingface.co/LiquidAI/LFM2.5-8B-A1B/blob/5dd22602c2e9f6a097b1de4c4efe0658b605015c/README.md\nLicensed under LFM Open License v1.0: https://huggingface.co/LiquidAI/LFM2.5-8B-A1B/blob/5dd22602c2e9f6a097b1de4c4efe0658b605015c/LICENSE\nModified for whatmodel.sh: selected benchmark cells, normalized percentages to fractions, and clarified metric/setup labels. No endorsement. Commercial reuse remains subject to the licence’s annual-revenue condition and other terms.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "The explicitly linked LFM Open License v1.0 includes documentation and grants reproduction, adaptation, display and distribution subject to all its terms. Section 5 imposes commercial revenue and Legal Entity conditions; each recipient must assess eligibility.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice.",
    "conditions": "LFM Open License v1.0 includes commercial revenue and Legal Entity conditions. Downstream users must assess their own eligibility; this compilation does not remove these restrictions."
  },
  {
    "id": "card-mistralai-devstral-small-2-24b-instruct-2512",
    "name": "mistralai/Devstral-Small-2-24B-Instruct-2512 model card",
    "match": [
      "huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/blob/55c5b41e98c2dbd21b0c8afffc540dcfc9eb5128/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "mistralai/Devstral-Small-2-24B-Instruct-2512 — mistralai/Devstral-Small-2-24B-Instruct-2512 model card\nSource: https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/blob/55c5b41e98c2dbd21b0c8afffc540dcfc9eb5128/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-mistralai-ministral-3-14b-instruct-2512",
    "name": "mistralai/Ministral-3-14B-Instruct-2512 model card",
    "match": [
      "huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "attribution": "mistralai/Ministral-3-14B-Instruct-2512 — mistralai/Ministral-3-14B-Instruct-2512 model card\nSource: https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction. Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Instruct table, rows Ministral 3 14B/8B/3B (Instruct 2512) on Arena Hard and MATH Maj@1.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-mistralai-ministral-3-14b-reasoning-2512",
    "name": "mistralai/Ministral-3-14B-Reasoning-2512 model card",
    "match": [
      "huggingface.co/mistralai/Ministral-3-14B-Reasoning-2512/blob/51f9210f3cd20f3452a80d5819d15dc61cc50630/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "mistralai/Ministral-3-14B-Reasoning-2512 — mistralai/Ministral-3-14B-Reasoning-2512 model card\nSource: https://huggingface.co/mistralai/Ministral-3-14B-Reasoning-2512/blob/51f9210f3cd20f3452a80d5819d15dc61cc50630/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-mistralai-ministral-3-3b-instruct-2512",
    "name": "mistralai/Ministral-3-3B-Instruct-2512 model card",
    "match": [
      "huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "mistralai/Ministral-3-3B-Instruct-2512 — mistralai/Ministral-3-3B-Instruct-2512 model card\nSource: https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-mistralai-ministral-3-3b-reasoning-2512",
    "name": "mistralai/Ministral-3-3B-Reasoning-2512 model card",
    "match": [
      "huggingface.co/mistralai/Ministral-3-3B-Reasoning-2512/blob/4a36357c811bf511a7b625d132e12f22408aac91/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "attribution": "mistralai/Ministral-3-3B-Reasoning-2512 — mistralai/Ministral-3-3B-Reasoning-2512 model card\nSource: https://huggingface.co/mistralai/Ministral-3-3B-Reasoning-2512/blob/4a36357c811bf511a7b625d132e12f22408aac91/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction. Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-mistralai-ministral-3-8b-instruct-2512",
    "name": "mistralai/Ministral-3-8B-Instruct-2512 model card",
    "match": [
      "huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "attribution": "mistralai/Ministral-3-8B-Instruct-2512 — mistralai/Ministral-3-8B-Instruct-2512 model card\nSource: https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction. Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-mistralai-ministral-3-8b-reasoning-2512",
    "name": "mistralai/Ministral-3-8B-Reasoning-2512 model card",
    "match": [
      "huggingface.co/mistralai/Ministral-3-8B-Reasoning-2512/blob/81eaece1948f3875421d9a45bc55487d10e2d894/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "mistralai/Ministral-3-8B-Reasoning-2512 — mistralai/Ministral-3-8B-Reasoning-2512 model card\nSource: https://huggingface.co/mistralai/Ministral-3-8B-Reasoning-2512/blob/81eaece1948f3875421d9a45bc55487d10e2d894/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-mistralai-mistral-small-4-119b-2603",
    "name": "mistralai/Mistral-Small-4-119B-2603 model card",
    "match": [
      "huggingface.co/mistralai/Mistral-Small-4-119B-2603/blob/a11f36bebf709121056b1dbcc943d1c6afbe494d/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "mistralai/Mistral-Small-4-119B-2603 — mistralai/Mistral-Small-4-119B-2603 model card\nSource: https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/blob/a11f36bebf709121056b1dbcc943d1c6afbe494d/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to mistralai. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "card-qwen-qwen3-coder-next",
    "name": "Qwen/Qwen3-Coder-Next model card",
    "match": [
      "huggingface.co/Qwen/Qwen3-Coder-Next/blob/a7fbcb5c0e12d62a448eaa0e260346bf5dcc0feb/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://www.apache.org/licenses/LICENSE-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "attribution": "Qwen/Qwen3-Coder-Next — Qwen/Qwen3-Coder-Next model card\nSource: https://huggingface.co/Qwen/Qwen3-Coder-Next/blob/a7fbcb5c0e12d62a448eaa0e260346bf5dcc0feb/README.md\nLicence declaration: Apache-2.0 (pinned README metadata).\nChanges: Selected own-model table cells or printed chart labels transcribed into JSON; percentages divided by 100, already-fractional values preserved; benchmark aliases/settings documented. Logos and full chart artwork are not redistributed in the results.\nAttribution to Qwen. No endorsement.",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Accepted as publisher-authored data in the licensed repository README (or the publisher’s own chart embedded in that README). The pinned README declares license: apache-2.0 through the Hub’s documented repository-licensing mechanism. Apache 2.0 section 1 includes documentation source; section 2 grants reproduction, derivative-work and distribution rights; section 4 conditions are retained. No separate documentation exclusion or incompatible notice was found for the selected own-model material. This is a repository-specific scope assessment, not a rule that first-party facts or linked external blogs are automatically reusable. A separate root LICENSE file is not required to recognize the explicit standard-licence declaration; the complete Apache text accompanies this extraction.",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "report-2511.23404v1",
    "name": "LFM2 Technical Report — Liquid AI own SmolLM3 evaluation",
    "match": [
      "arxiv.org/html/2511.23404v1"
    ],
    "license": "CC BY 4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "attribution": "Liquid AI (2025). LFM2 Technical Report. arXiv:2511.23404v1. https://arxiv.org/html/2511.23404v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nSelected Table7 SmolLM3-3B own reevaluation cells and AppendixC methods; percentages normalized to fractions and benchmark names clarified. No endorsement.",
    "exact": true,
    "kind": "paper",
    "icon": "arxiv.org",
    "license_url": "/licenses/cards-gap-NOTICE.txt",
    "scope": "Pinned report explicitly declares CC BY4.0; Table7 states all results use the authors’ internal evaluation harness. These are Liquid AI’s independent evaluations of SmolLM3-3B, not third-party copied comparisons. The report licence permits commercial reuse independently of any LFM model licence/revenue threshold. Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Table 7 (internal harness, all columns): SmolLM3-3B plus LFM2-8B-A1B (lab), Llama-3.2-3B-Instruct, Gemma-3-4B-it and Qwen3-4B (original). Pass 12 (2026-10-11): also the LFM2-2.6B column of Table 7 (same internal harness; by: lab).",
    "quote": "Explicit licence on the pinned publication. Full terms, attribution and modification notices retained in the linked notice."
  },
  {
    "id": "speech-amharic-addis-bysa",
    "name": "Addis AI Amharic TTS benchmark",
    "kind": "dataset",
    "license": "CC BY-SA 4.0",
    "terms_url": "https://huggingface.co/datasets/addisai/amharic-tts-benchmark/blob/a60556cd62dd2fbfce83a09c11b088e1e027ce64/README.md#licensing",
    "checked": "2026-09-25",
    "attribution": "Amharic TTS Benchmark by Addis AI (2026-08-12), CC BY-SA 4.0. Source: addisai/amharic-tts-benchmark. Changes: select the 87-prompt primary CER results; map explicit system IDs; independently recompute edit counts; retain original fractional units. This numerical adaptation is CC BY-SA 4.0. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/datasets/addisai/amharic-tts-benchmark/blob/a60556cd62dd2fbfce83a09c11b088e1e027ce64/data/summary.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "README Licensing table expressly grants CC BY-SA 4.0 for data/*.jsonl, data/summary.json, scorer, normalization code and dataset card. No audio reused.",
    "license_url": "/licenses/speech-followup-NOTICES.txt"
  },
  {
    "id": "speech-vibevoice-mit-figure",
    "name": "Microsoft VibeVoice own subjective evaluation figure",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://github.com/microsoft/VibeVoice/blob/1541f590c7099820f10ea012f48d2399282df69f/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2025 Microsoft. VibeVoice repository, MIT License. Exact labelled preference MOS values transcribed from Figures/VibeVoice-TTS-results.jpg. Changes: selected three model rows and normalized (MOS-1)/4 for a unit-interval higher-better score. Retain MIT notice. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/microsoft/VibeVoice/blob/1541f590c7099820f10ea012f48d2399282df69f/Figures/VibeVoice-TTS-results.jpg"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Microsoft-owned evaluation figure and associated documentation distributed in the MIT repository. Not a licence inference from model weights or from third-party copies.",
    "license_url": "/licenses/speech-followup-NOTICES.txt"
  },
  {
    "id": "speech-vapi-humanness-20260925",
    "name": "The Humanness Index by Vapi · September 2026 public standings",
    "kind": "leaderboard",
    "license": "CC BY 4.0",
    "terms_url": "https://github.com/VapiAI/humanness-index/blob/e2b8f8653ed491a9f6038de98d6ef999c2729743/README.md#licensing",
    "checked": "2026-09-25",
    "attribution": "The Humanness Index™ by Vapi. Numerical vote data and standings, CC BY 4.0. https://humannessindex.vapi.ai/api/models. Changes: selected exact model rows, mapped documented model identities, retained raw published ratings and counts. No endorsement. Adapted compilation CC BY-SA 4.0; upstream grant retained.",
    "verdict": "allowed",
    "match": [
      "humannessindex.vapi.ai/api/models"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "README expressly licenses numerical vote data and standings CC BY 4.0. Code/docs Apache-2.0; audio and logos excluded from this extraction.",
    "license_url": "/licenses/speech-followup-NOTICES.txt"
  },
  {
    "id": "speech-vapi-humanness-20260610",
    "name": "The Humanness Index by Vapi · June 2026 production export",
    "kind": "leaderboard",
    "license": "CC BY 4.0",
    "terms_url": "https://github.com/VapiAI/humanness-index/blob/e2b8f8653ed491a9f6038de98d6ef999c2729743/README.md#licensing",
    "checked": "2026-09-25",
    "attribution": "The Humanness Index™ by Vapi. Numerical vote data and standings, CC BY 4.0. https://github.com/VapiAI/humanness-index/blob/e2b8f8653ed491a9f6038de98d6ef999c2729743/src/server/seed-standings.json. Changes: selected exact model rows, mapped documented model identities, retained raw published ratings and counts. No endorsement. Adapted compilation CC BY-SA 4.0; upstream grant retained.",
    "verdict": "allowed",
    "match": [
      "github.com/VapiAI/humanness-index/blob/e2b8f8653ed491a9f6038de98d6ef999c2729743/src/server/seed-standings.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "README expressly licenses numerical vote data and standings CC BY 4.0. Code/docs Apache-2.0; audio and logos excluded from this extraction.",
    "license_url": "/licenses/speech-followup-NOTICES.txt"
  },
  {
    "id": "speech-granite41-punctuation-apache",
    "name": "IBM Granite Speech 4.1 own punctuation evaluation",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/ibm-granite/granite-speech-4.1-2b/blob/de575db64086f84fdc79da4932d1076e965bc546/README.md",
    "checked": "2026-09-25",
    "attribution": "IBM Granite team, Granite Speech 4.1 2B model card, Apache License 2.0. Changes: selected the LScln punctuation-error result, divided percent by 100, retained Cap-F1 as companion metadata. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/ibm-granite/granite-speech-4.1-2b/blob/de575db64086f84fdc79da4932d1076e965bc546/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "First-party README numerical evaluation table. Repository/card metadata and an explicit License section both declare Apache 2.0; no distinct restriction on the own-results table. No licence inferred for copied third-party leaderboard charts.",
    "license_url": "/licenses/speech-followup-NOTICES.txt"
  },
  {
    "id": "visual-gap-amd-mxfp",
    "name": "Breaking the Accuracy-Speed Barrier: How MXFP4/6 Quantization Revolutionizes Image and Video Generation",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://github.com/ROCm/rocm-blogs/blob/b3bf907bd73c596c49c77b1d178c7c41bcd2713c/blogs/artificial-intelligence/mxfp-t2i-t2v/README.md#L28-L48",
    "checked": "2026-09-25",
    "attribution": "Hongyi Yao, Zhe Li, Han Wang, Xuanwu Yin, Dong Li, Steve Reinhardt, Ephrem Wu, Emad Barsoum; Copyright (c) 2025 Advanced Micro Devices, Inc. (AMD)",
    "verdict": "allowed",
    "match": [
      "github.com/ROCm/rocm-blogs/blob/b3bf907bd73c596c49c77b1d178c7c41bcd2713c/blogs/artificial-intelligence/mxfp-t2i-t2v/README.md"
    ],
    "exact": true,
    "license_url": "/licenses/visual-gap-amd-mxfp.txt",
    "icon": "huggingface.co",
    "quote": "Only the retained original measurements and documentation within the explicitly licensed artifact. No licence inferred for external leaderboards, copied third-party scores, model weights, generated media, or benchmark inputs."
  },
  {
    "id": "visual-gap-jenga",
    "name": "Training-Free Efficient Video Generation via Dynamic Token Carving",
    "kind": "paper",
    "license": "CC-BY-SA-4.0",
    "terms_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "checked": "2026-09-25",
    "attribution": "Yuechen Zhang , Jinbo Xing , Bin Xia , Shaoteng Liu , Bohao Peng , Xin Tao , Pengfei Wan , Eric Lo , Jiaya Jia",
    "verdict": "allowed",
    "match": [
      "arxiv.org/pdf/2505.16864v2"
    ],
    "exact": true,
    "license_url": "/licenses/visual-gap-jenga.txt",
    "icon": "arxiv.org",
    "quote": "Only the retained original measurements and documentation within the explicitly licensed artifact. No licence inferred for external leaderboards, copied third-party scores, model weights, generated media, or benchmark inputs."
  },
  {
    "id": "visual-gap-mobilewan",
    "name": "MobileWan: Closing the Quality Gap for Mobile Video Diffusion",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "checked": "2026-09-25",
    "attribution": "Mohsen Ghafoorian , Denis Korzhenkov , Adil Karjauv , Ioannis Lelekas , Noor Fathima , Spyridon Stasis , Hanno Ackermann , Boris van Breugel , Markus Nagel , Fatih Porikli , Animesh Karnewar , Amirhossein Habibian",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2607.06173v1"
    ],
    "exact": true,
    "license_url": "/licenses/visual-gap-mobilewan.txt",
    "icon": "arxiv.org",
    "quote": "Only the retained original measurements and documentation within the explicitly licensed artifact. No licence inferred for external leaderboards, copied third-party scores, model weights, generated media, or benchmark inputs."
  },
  {
    "id": "visual-gap-target-bench",
    "name": "Target-Bench: Can Video World Models Achieve Mapless Path Planning with Semantic Targets?",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "checked": "2026-09-25",
    "attribution": "Dingrui Wang , Zhihao Liang , Hongyuan Ye , Zhexiao Sun , Zhaowei Lu , Yuchen Zhang , Yuyu Zhao , Yuan Gao , Marvin Seegert , Finn Schäfer , Haotong Qin , Wei Li , Luigi Palmieri , Felix Jahncke , Mattia Piccinini , Johannes Betz",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2511.17792v2"
    ],
    "exact": true,
    "license_url": "/licenses/visual-gap-target-bench.txt",
    "icon": "arxiv.org",
    "quote": "Only the retained original measurements and documentation within the explicitly licensed artifact. No licence inferred for external leaderboards, copied third-party scores, model weights, generated media, or benchmark inputs."
  },
  {
    "id": "bertrande-direct-provider-observations",
    "name": "Bertrande direct-provider price observations",
    "license": "CC-BY-4.0",
    "attribution": "Bertrande (Paul Roulleau), https://bertrande.com. CC BY 4.0. Each derived row retains its service URL and observation date; fields selected and units normalized. No endorsement.",
    "kind": "dataset",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "bertrande.com/en/ai/deepinfra/ibm-granite--granite-4.2-30b",
      "bertrande.com/en/ai/deepinfra/ibm-granite--granite-4.2-3b",
      "bertrande.com/en/ai/replicate/alibaba--happyhorse-1.0",
      "bertrande.com/en/ai/replicate/black-forest-labs--flux-dev",
      "bertrande.com/en/ai/replicate/black-forest-labs--flux-schnell",
      "bertrande.com/en/ai/replicate/inworld--realtime-tts-2"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Explicit grant for publisher-authored downloadable observation catalogue, not an inferred grant for upstream provider pages. Only selected direct-provider records from source levels 1 or 3 are reused. No AA, OpenRouter, router performance or aggregator-derived observations.",
    "terms_url": "https://bertrande.com/en/data/"
  },
  {
    "id": "gap-card-qwen3-5-0-8b",
    "name": "Qwen/Qwen3.5-0.8B official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3.5-0.8B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-0.8B/blob/2fc06364715b967f1860aea9cf38778875588b17/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-0.8B/resolve/2fc06364715b967f1860aea9cf38778875588b17/README.md"
  },
  {
    "id": "gap-card-qwen3-5-2b",
    "name": "Qwen/Qwen3.5-2B official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3.5-2B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-2B/blob/15852e8c16360a2fea060d615a32b45270f8a8fc/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-2B/resolve/15852e8c16360a2fea060d615a32b45270f8a8fc/README.md"
  },
  {
    "id": "gap-card-qwen3-5-4b",
    "name": "Qwen/Qwen3.5-4B official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3.5-4B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3.5-4B/blob/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3.5-4B/resolve/851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a/README.md"
  },
  {
    "id": "gap-card-qwen2-5-7b",
    "name": "Qwen/Qwen2.5-7B-Instruct official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen2.5-7B-Instruct and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/a09a35458c702b33eeacc393d103063234e8bc28/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/resolve/a09a35458c702b33eeacc393d103063234e8bc28/README.md"
  },
  {
    "id": "gap-card-qwen2-5-14b",
    "name": "Qwen/Qwen2.5-14B-Instruct official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen2.5-14B-Instruct and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen2.5-14B-Instruct/blob/cf98f3b3bbb457ad9e2bb7baf9a0125b6b88caa8/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen2.5-14B-Instruct/resolve/cf98f3b3bbb457ad9e2bb7baf9a0125b6b88caa8/README.md"
  },
  {
    "id": "gap-card-qwen2-5-32b",
    "name": "Qwen/Qwen2.5-32B-Instruct official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen2.5-32B-Instruct and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen2.5-32B-Instruct/blob/5ede1c97bbab6ce5cda5812749b4c0bdf79b18dd/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen2.5-32B-Instruct/resolve/5ede1c97bbab6ce5cda5812749b4c0bdf79b18dd/README.md"
  },
  {
    "id": "gap-card-qwen2-5-coder-7b",
    "name": "Qwen/Qwen2.5-Coder-7B-Instruct official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen2.5-Coder-7B-Instruct and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct/blob/c03e6d358207e414f1eca0bb1891e29f1db0e242/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct/resolve/c03e6d358207e414f1eca0bb1891e29f1db0e242/README.md"
  },
  {
    "id": "gap-card-qwen2-5-coder-32b",
    "name": "Qwen/Qwen2.5-Coder-32B-Instruct official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen2.5-Coder-32B-Instruct and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct/blob/381fc969f78efac66bc87ff7ddeadb7e73c218a7/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct/resolve/381fc969f78efac66bc87ff7ddeadb7e73c218a7/README.md"
  },
  {
    "id": "gap-card-qwen3-0-6b",
    "name": "Qwen/Qwen3-0.6B official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3-0.6B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3-0.6B/blob/c1899de289a04d12100db370d81485cdf75e47ca/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-0.6B/resolve/c1899de289a04d12100db370d81485cdf75e47ca/README.md"
  },
  {
    "id": "gap-card-qwen3-1-7b",
    "name": "Qwen/Qwen3-1.7B official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3-1.7B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3-1.7B/blob/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-1.7B/resolve/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/README.md"
  },
  {
    "id": "gap-card-qwen3-8b",
    "name": "Qwen/Qwen3-8B official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3-8B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3-8B/blob/b968826d9c46dd6066d109eabc6255188de91218/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-8B/resolve/b968826d9c46dd6066d109eabc6255188de91218/README.md"
  },
  {
    "id": "gap-card-qwen3-14b",
    "name": "Qwen/Qwen3-14B official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3-14B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3-14B/blob/40c069824f4251a91eefaf281ebe4c544efd3e18/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-14B/resolve/40c069824f4251a91eefaf281ebe4c544efd3e18/README.md"
  },
  {
    "id": "gap-card-qwen3-4b-instruct-2507",
    "name": "Qwen/Qwen3-4B-Instruct-2507 official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3-4B-Instruct-2507 and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3-4B-Instruct-2507/blob/cdbee75f17c01a7cc42f958dc650907174af0554/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507/resolve/cdbee75f17c01a7cc42f958dc650907174af0554/README.md"
  },
  {
    "id": "gap-card-qwen3-4b-thinking-2507",
    "name": "Qwen/Qwen3-4B-Thinking-2507 official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3-4B-Thinking-2507 and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3-4B-Thinking-2507/blob/768f209d9ea81521153ed38c47d515654e938aea/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507/resolve/768f209d9ea81521153ed38c47d515654e938aea/README.md"
  },
  {
    "id": "gap-card-qwen3-30b-a3b-thinking-2507",
    "name": "Qwen/Qwen3-30B-A3B-Thinking-2507 official repository context",
    "license": "Apache-2.0",
    "attribution": "Qwen/Qwen3-30B-A3B-Thinking-2507 and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/blob/144afc2f379b542fdd4e85a1fcd5e1f79112d95d/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/resolve/144afc2f379b542fdd4e85a1fcd5e1f79112d95d/README.md"
  },
  {
    "id": "gap-card-granite-4-0-micro",
    "name": "ibm-granite/granite-4.0-micro official repository context",
    "license": "Apache-2.0",
    "attribution": "ibm-granite/granite-4.0-micro and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/ibm-granite/granite-4.0-micro/blob/56111ae135df9c53a78c99028e7bc24035a9e979/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/ibm-granite/granite-4.0-micro/resolve/56111ae135df9c53a78c99028e7bc24035a9e979/README.md"
  },
  {
    "id": "gap-card-granite-4-0-h-small",
    "name": "ibm-granite/granite-4.0-h-small official repository context",
    "license": "Apache-2.0",
    "attribution": "ibm-granite/granite-4.0-h-small and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/ibm-granite/granite-4.0-h-small/blob/b8c0982bab7fde4eb48110f5a069527c008fab39/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/ibm-granite/granite-4.0-h-small/resolve/b8c0982bab7fde4eb48110f5a069527c008fab39/README.md"
  },
  {
    "id": "gap-card-smollm3-3b",
    "name": "HuggingFaceTB/SmolLM3-3B official repository context",
    "license": "Apache-2.0",
    "attribution": "HuggingFaceTB/SmolLM3-3B and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/HuggingFaceTB/SmolLM3-3B/blob/a07cc9a04f16550a088caea529712d1d335b0ac1/README.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B/resolve/a07cc9a04f16550a088caea529712d1d335b0ac1/README.md"
  },
  {
    "id": "gap-card-seed-oss-36b",
    "name": "ByteDance-Seed/Seed-OSS-36B-Instruct official repository context",
    "license": "Apache-2.0",
    "attribution": "ByteDance-Seed/Seed-OSS-36B-Instruct and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/blob/497f1dca95ebdec98e41d517b9f060ee753c902f/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/resolve/497f1dca95ebdec98e41d517b9f060ee753c902f/README.md"
  },
  {
    "id": "gap-card-phi-4-reasoning",
    "name": "microsoft/Phi-4-reasoning official repository context",
    "license": "MIT",
    "attribution": "microsoft/Phi-4-reasoning and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/microsoft/Phi-4-reasoning/blob/1de18ec97600877ce63dbf60c73b998da99f0195/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/microsoft/Phi-4-reasoning/resolve/1de18ec97600877ce63dbf60c73b998da99f0195/README.md"
  },
  {
    "id": "gap-card-olmo-3-1-32b-think",
    "name": "allenai/Olmo-3.1-32B-Think official repository context",
    "license": "Apache-2.0",
    "attribution": "allenai/Olmo-3.1-32B-Think and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/allenai/Olmo-3.1-32B-Think/blob/832c3f543499af8fe68b88359501de9cb7840544/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think/resolve/832c3f543499af8fe68b88359501de9cb7840544/README.md"
  },
  {
    "id": "gap-card-olmo-3-7b-think",
    "name": "allenai/Olmo-3-7B-Think official repository context",
    "license": "Apache-2.0",
    "attribution": "allenai/Olmo-3-7B-Think and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/allenai/Olmo-3-7B-Think/blob/d97e442d7cc678210054dbcc9b440894d62c89a4/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/allenai/Olmo-3-7B-Think/resolve/d97e442d7cc678210054dbcc9b440894d62c89a4/README.md"
  },
  {
    "id": "gap-card-olmo-3-7b-instruct",
    "name": "allenai/Olmo-3-7B-Instruct official repository context",
    "license": "Apache-2.0",
    "attribution": "allenai/Olmo-3-7B-Instruct and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/allenai/Olmo-3-7B-Instruct/blob/6e5971d9eba42665f5bd5a0fcf047f299ce1dccc/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/allenai/Olmo-3-7B-Instruct/resolve/6e5971d9eba42665f5bd5a0fcf047f299ce1dccc/README.md"
  },
  {
    "id": "gap-card-ministral-3-3b",
    "name": "mistralai/Ministral-3-3B-Instruct-2512 official repository context",
    "license": "Apache-2.0",
    "attribution": "mistralai/Ministral-3-3B-Instruct-2512 and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/resolve/b35d4dfe56c142746f54dbd64f579faab2744308/README.md"
  },
  {
    "id": "gap-card-ministral-3-8b",
    "name": "mistralai/Ministral-3-8B-Instruct-2512 official repository context",
    "license": "Apache-2.0",
    "attribution": "mistralai/Ministral-3-8B-Instruct-2512 and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/resolve/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/README.md"
  },
  {
    "id": "gap-card-ministral-3-14b",
    "name": "mistralai/Ministral-3-14B-Instruct-2512 official repository context",
    "license": "Apache-2.0",
    "attribution": "mistralai/Ministral-3-14B-Instruct-2512 and contributors. Context field extracted from the pinned repository; converted to decimal k. No benchmark cells copied.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/config.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Pinned model repository explicitly declares the standard licence for repository content. Apache Source form includes documentation and configuration. Reviewed selected own-model context fields; linked third-party benchmark data is not reused. Full licence text and available upstream notices retained.",
    "terms_url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/resolve/29439f81c2be264d8d393273f99e7db9c0961120/README.md"
  },
  {
    "id": "gap-google-gemini31",
    "name": "Google Gemini 3.1 Pro preview API model card",
    "license": "CC-BY-4.0",
    "attribution": "Google, model documentation, CC BY 4.0, retrieved 25 September 2026. Selected context facts; model IDs mapped. No endorsement.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/prices-gap-NOTICE.txt",
    "quote": "Explicit CC BY 4.0 footer for page content, not an inference from the model weight licence. Only context/identity text is selected; no benchmark tables.",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "gap-google-gemma3",
    "name": "Google DeepMind Gemma 3 270M IT model card",
    "license": "CC-BY-4.0",
    "attribution": "Google, model documentation, CC BY 4.0, retrieved 25 September 2026. Selected context facts; model IDs mapped. No endorsement.\nGoogle DeepMind / Gemma Team. Google DeepMind Gemma 3 270M IT model card.\nSource: https://ai.google.dev/gemma/docs/core/model_card_3\nLicence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "ai.google.dev/gemma/docs/core/model_card_3"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "The Google developer model-card page expressly licenses its page content under CC BY4.0 except otherwise noted; no different licence is marked on the selected benchmark tables. The model Terms link governs model use and does not cancel the explicit page-content grant. Data is extracted from this CC-BY document. The pinned HF card is a matching evidence mirror, not claimed to be relicensed from its Gemma metadata. Context metadata in this document is also reviewed.",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "gap-google-gemma3n",
    "name": "Google DeepMind Gemma 3n E2B IT model card",
    "license": "CC-BY-4.0",
    "attribution": "Google, model documentation, CC BY 4.0, retrieved 25 September 2026. Selected context facts; model IDs mapped. No endorsement.\nGoogle DeepMind / Gemma Team. Google DeepMind Gemma 3n E2B IT model card.\nSource: https://ai.google.dev/gemma/docs/gemma-3n/model_card\nLicence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement.",
    "kind": "first-party",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "match": [
      "ai.google.dev/gemma/docs/gemma-3n/model_card"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "The Google developer model-card page expressly licenses its page content under CC BY4.0 except otherwise noted; no different licence is marked on the selected benchmark tables. The model Terms link governs model use and does not cancel the explicit page-content grant. Data is extracted from this CC-BY document. The pinned HF card is a matching evidence mirror, not claimed to be relicensed from its Gemma metadata. Context metadata in this document is also reviewed.",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/"
  },
  {
    "id": "card-ernie-45-posttrained-final",
    "name": "ERNIE 4.5 post-trained 300B-A47B benchmark chart",
    "match": [
      "github.com/PaddlePaddle/ERNIE/blob/790a50b045d1aca2753d5395d8bec0806b2e6925/README.md"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://github.com/PaddlePaddle/ERNIE/blob/790a50b045d1aca2753d5395d8bec0806b2e6925/LICENSE",
    "checked": "2026-09-25",
    "attribution": "PaddlePaddle Authors / Baidu. ERNIE 4.5 post-trained 300B-A47B benchmark chart.\nSource: https://github.com/PaddlePaddle/ERNIE/blob/790a50b045d1aca2753d5395d8bec0806b2e6925/README.md\nLicence: Apache-2.0 — https://github.com/PaddlePaddle/ERNIE/blob/790a50b045d1aca2753d5395d8bec0806b2e6925/LICENSE\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement.",
    "verdict": "allowed",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "Root Apache2 licence covers repository documentation, including the publisher-owned benchmark chart embedded in the post-trained section. Selected numerical labels are Baidu’s own results. The external image remains content-hashed rather than falsely commit-pinned. No scope exclusion for this author-owned material was found."
  },
  {
    "id": "card-final-ling-3-0-flash",
    "name": "inclusionAI/Ling-3.0-flash model card",
    "match": [
      "huggingface.co/inclusionAI/Ling-3.0-flash/blob/ef06d91fe382109ae82647da88ff99b0f11745b0/README.md"
    ],
    "license": "MIT",
    "terms_url": "https://raw.githubusercontent.com/spdx/license-list-data/31ba1a50e5397e00a304dbadc76531740e89ee48/text/MIT.txt",
    "checked": "2026-09-25",
    "attribution": "inclusionAI. inclusionAI/Ling-3.0-flash model card.\nSource: https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/ef06d91fe382109ae82647da88ff99b0f11745b0/README.md\nLicence: MIT — https://raw.githubusercontent.com/spdx/license-list-data/31ba1a50e5397e00a304dbadc76531740e89ee48/text/MIT.txt\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement.",
    "verdict": "allowed",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "Pinned publisher repository README explicitly declares license: mit. The standard MIT grant expressly covers associated documentation and permits commercial copying/publication/distribution subject to retention of the permission and copyright notices. No different documentation licence or exception was found. No separate root licence file is required to recognize this explicit declaration. Only author-evaluated model results embedded in the README are selected; linked blogs and competitor results are not presumed licensed."
  },
  {
    "id": "card-final-apriel-1-6-15b-thinker",
    "name": "ServiceNow-AI/Apriel-1.6-15b-Thinker model card",
    "match": [
      "huggingface.co/ServiceNow-AI/Apriel-1.6-15b-Thinker/blob/5810d5f1723032300131e677450d7f95a15a05e7/README.md"
    ],
    "license": "MIT",
    "terms_url": "https://raw.githubusercontent.com/spdx/license-list-data/31ba1a50e5397e00a304dbadc76531740e89ee48/text/MIT.txt",
    "checked": "2026-10-06",
    "attribution": "ServiceNow-AI. ServiceNow-AI/Apriel-1.6-15b-Thinker model card.\nSource: https://huggingface.co/ServiceNow-AI/Apriel-1.6-15b-Thinker/blob/5810d5f1723032300131e677450d7f95a15a05e7/README.md\nLicence: MIT — https://raw.githubusercontent.com/spdx/license-list-data/31ba1a50e5397e00a304dbadc76531740e89ee48/text/MIT.txt\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement. Added internally evaluated comparison columns from the same five rows; source statement and effort labels preserved.",
    "verdict": "allowed",
    "exact": true,
    "kind": "first-party",
    "icon": "huggingface.co",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "Pinned publisher repository README explicitly declares license: mit. The standard MIT grant expressly covers associated documentation and permits commercial copying/publication/distribution subject to retention of the permission and copyright notices. No different documentation licence or exception was found. No separate root licence file is required to recognize this explicit declaration. Only author-evaluated model results embedded in the README are selected; linked blogs and competitor results are not presumed licensed. Scope clarification 2026-10-06: the card explicitly states that all benchmarks outside the AA Index v3.0 were evaluated internally. The five already-approved internal benchmark rows also contain original comparison-model runs; those exact columns are now included. This does not approve any copied AA benchmark, index, projected score, or external blog."
  },
  {
    "id": "report-marco-moe-260425578v1",
    "name": "Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling",
    "match": [
      "arxiv.org/html/2604.25578v1"
    ],
    "license": "CC BY 4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "checked": "2026-10-06",
    "attribution": "Fan Jiang, Yu Zhao, Chenyang Lyu, Tianqi Shi, Yichao Du, Feihu Jiang, Longyue Wang and Weihua Luo (2026). Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling.\nSource: https://arxiv.org/html/2604.25578v1\nLicence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement.",
    "verdict": "allowed",
    "exact": true,
    "kind": "paper",
    "icon": "arxiv.org",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "Pinned report explicitly CC BY 4.0 (arXiv abs page licence link http://creativecommons.org/licenses/by/4.0/). Section 4.3.1 says all Instruct models, including every baseline, were evaluated by the authors with the Light-Eval framework (temperature 1.0, top_p 1.0, 8,192 max tokens; Avg@8, Acc@1 for GlobalMMLU/MMMLU). Selected Table 11 cells for LFM2-24B-A2B, Qwen3-1.7B (non-thinking), Qwen3-4B-Instruct-2507, Ministral-3-3B/8B-Instruct-2512, Gemma3-12B-Instruct and Granite-4.0-H-Small are independent measurements of those released checkpoints. The model-weight licences (LFM Open License, Gemma terms, etc.) do not attach to this CC BY article."
  },
  {
    "id": "acl-konubench-2026",
    "name": "Thunder-KoNUBench: A Corpus-Aligned Benchmark for Korean Negation Understanding",
    "match": [
      "aclanthology.org/2026.findings-acl.324.pdf"
    ],
    "license": "CC BY 4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "checked": "2026-09-25",
    "attribution": "Sungmok Jung, Yeonkyoung So, Joonhak Lee, Sangho Kim, Yelim Ahn and Jaejin Lee; ©2026 Association for Computational Linguistics. Thunder-KoNUBench: A Corpus-Aligned Benchmark for Korean Negation Understanding.\nSource: https://aclanthology.org/2026.findings-acl.324.pdf\nLicence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement.",
    "verdict": "allowed",
    "exact": true,
    "kind": "paper",
    "icon": "arxiv.org",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "This is an ACL-owned 2026 publication, explicitly covered by ACL’s post2016 CC BY4.0 publication policy. Selected table statistics are the paper authors’ original evaluations, not reprinted LG card scores. EXAONE’s NC model licence and any NC benchmark dataset/code licence are not transferred to the article’s independently authored results. Only aggregate numerical findings and methods are extracted, not dataset questions, rubric text, inference outputs or code. No alternate licence is marked on these numerical tables."
  },
  {
    "id": "acl-kcl-2026",
    "name": "Korean Canonical Legal Benchmark: Toward Knowledge-Independent Evaluation of LLMs’ Legal Reasoning Capabilities",
    "match": [
      "aclanthology.org/2026.eacl-short.17.pdf"
    ],
    "license": "CC BY 4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "checked": "2026-10-09",
    "attribution": "Hongseok Oh, Wonseok Hwang and Kyoung-Woon On; ©2026 Association for Computational Linguistics. Korean Canonical Legal Benchmark: Toward Knowledge-Independent Evaluation of LLMs’ Legal Reasoning Capabilities.\nSource: https://aclanthology.org/2026.eacl-short.17.pdf\nLicence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/\nChanges: Selected numerical cells transcribed into JSON; percentages divided by100; benchmark names and settings clarified. No model weights, test questions, full chart artwork or model-generated answers redistributed.\nNo endorsement.",
    "verdict": "allowed",
    "exact": true,
    "kind": "paper",
    "icon": "arxiv.org",
    "license_url": "/licenses/cards-final-NOTICE.txt",
    "quote": "This is an ACL-owned 2026 publication, explicitly covered by ACL’s post2016 CC BY4.0 publication policy. Selected table statistics are the paper authors’ original evaluations, not reprinted LG card scores. EXAONE’s NC model licence and any NC benchmark dataset/code licence are not transferred to the article’s independently authored results. Only aggregate numerical findings and methods are extracted, not dataset questions, rubric text, inference outputs or code. No alternate licence is marked on these numerical tables.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Tables 2 and 3, vanilla column, every catalogued model with a pinned checkpoint (23 models), not only Exaone 4.0 32B."
  },
  {
    "id": "first-party-meta",
    "name": "Meta official pricing and model documentation",
    "match": [
      "dev.meta.ai/docs/pricing-rate-limits",
      "github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3_1/MODEL_CARD.md",
      "github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3_3/MODEL_CARD.md",
      "github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama4/MODEL_CARD.md",
      "arxiv.org/abs/2205.01068v4",
      "arxiv.org/abs/2308.12950v3",
      "github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama2/MODEL_CARD.md",
      "huggingface.co/codellama/CodeLlama-70b-hf/blob/cc11b0f79c8a072bd1e1f32ec280e1ff8ec018f6/README.md",
      "huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct/blob/9eb2daaa8597bf192a8b0e73f848f3a102794df5/README.md",
      "huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct/blob/e305d2a43a4adc6987308fe7d896fb8ec5f1a5d8/README.md",
      "huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct/blob/50fd307e57011801c7833c87efa1984ddf2db42f/README.md",
      "huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct/blob/8afb486c1db24fe5011ec46dfbe5b5dccdb575c2/README.md",
      "huggingface.co/codellama/CodeLlama-70b-Instruct-hf/blob/397cae981dffaf5d5c9c90e89a0a75a850528b70/README.md",
      "huggingface.co/codellama/CodeLlama-70b-Python-hf/blob/e037619532ec8449a01ecd5b253d065ac81637b6/README.md",
      "developer.meta.com/ai/models/muse-voice-transcribe"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://dev.meta.ai/docs/pricing-rate-limits",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "meta.ai",
    "attribution": "Meta official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-xai",
    "name": "xAI official pricing and model documentation",
    "match": [
      "docs.x.ai/developers/models.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.x.ai/developers/models.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "x.ai",
    "attribution": "xAI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-aws",
    "name": "Amazon Web Services official pricing and model documentation",
    "match": [
      "aws.amazon.com/bedrock/pricing",
      "aws.amazon.com/nova/pricing",
      "aws.amazon.com/polly/pricing",
      "docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html",
      "aws.amazon.com/about-aws/whats-new/2025/12/amazon-nova-2-omni-preview"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://aws.amazon.com/nova/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "aws.amazon.com",
    "attribution": "Amazon Web Services official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-cohere",
    "name": "Cohere official pricing and model documentation",
    "match": [
      "docs.cohere.com/docs/models",
      "huggingface.co/CohereLabs/aya-23-35B/blob/ce62acc9f088f423460cd42cd799f5cc53cc7c99/README.md",
      "huggingface.co/CohereLabs/aya-23-8B/blob/89da1a0ed02d6130f93ae0ffdbedb63b760c0471/README.md",
      "huggingface.co/CohereLabs/aya-expanse-32b/blob/b306ea27e360683b50c005d7fcbad6a242317910/README.md",
      "huggingface.co/CohereLabs/aya-expanse-8b/blob/5062468bf9bc0c6035fd64e06274333ec127d980/README.md",
      "huggingface.co/CohereLabs/aya-vision-32b/blob/0554d66834922fc0f2e5f47a12f78464f4a98533/README.md",
      "huggingface.co/CohereLabs/c4ai-command-r-08-2024/blob/dc835b893cd3fb8f14b24970dbc2a0a6d3c22ee3/README.md",
      "huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024/blob/e808c1a2249354ca211c9f08d1338e5039f633f8/README.md",
      "huggingface.co/CohereLabs/c4ai-command-r-plus/blob/3de12c88b4c6c882389a6f424cc2074383aa17d8/README.md",
      "huggingface.co/CohereLabs/c4ai-command-r-v01/blob/760ddb6c203d87ebdbe3c9785b49570e1bf95585/README.md",
      "huggingface.co/CohereLabs/c4ai-command-r7b-12-2024/blob/4f3d0aa6856e322f2f4480fe65420d5d53d297b8/README.md",
      "cohere.com/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.cohere.com/docs/models",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "cohere.com",
    "attribution": "Cohere official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-alibaba-cloud",
    "name": "Alibaba Cloud Model Studio official pricing and model documentation",
    "match": [
      "alibabacloud.com/help/en/model-studio/model-pricing",
      "alibabacloud.com/help/en/model-studio/qwen-image-3-0-pro",
      "alibabacloud.com/help/en/model-studio/text-generation-model",
      "alibabacloud.com/help/en/model-studio/model-qwen3-max",
      "alibabacloud.com/help/en/model-studio/qwen-plus",
      "alibabacloud.com/help/en/model-studio/qwen-omni"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "alibabacloud.com",
    "attribution": "Alibaba Cloud Model Studio official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-together",
    "name": "Together AI official pricing and model documentation",
    "match": [
      "docs.together.ai/docs/serverless/models.md",
      "together.ai/models/ideogram-40",
      "together.ai/models/sd-xl",
      "together.ai/pricing",
      "huggingface.co/togethercomputer/RedPajama-INCITE-7B-Base/blob/78f7e482443971f4873ba3239f0ac810a367833b/config.json",
      "huggingface.co/togethercomputer/RedPajama-INCITE-Base-3B-v1/blob/094fbdd0c911feb485ce55de1952ab2e75277e1e/config.json"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://www.together.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "together.ai",
    "attribution": "Together AI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-stepfun",
    "name": "StepFun official pricing and model documentation",
    "match": [
      "platform.stepfun.ai/docs/en/guides/models/step-3.7-flash.md",
      "platform.stepfun.ai/docs/en/guides/pricing/details.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://platform.stepfun.ai/docs/en/guides/pricing/details.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "stepfun.ai",
    "attribution": "StepFun official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-meituan-longcat",
    "name": "Meituan LongCat API Platform official pricing and model documentation",
    "match": [
      "longcat.chat/platform/docs/pricing/longcat-2.0"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://longcat.chat/platform/docs/pricing/longcat-2.0",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "longcat.chat",
    "attribution": "Meituan LongCat API Platform official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-deepinfra",
    "name": "DeepInfra official pricing and model documentation",
    "match": [
      "deepinfra.com/ByteDance/Seedream-4",
      "deepinfra.com/ByteDance/Seedream-5.0-Pro",
      "deepinfra.com/Qwen/Qwen3-ASR-1.7B",
      "deepinfra.com/Wan-AI/Wan2.6-T2I",
      "deepinfra.com/deepseek-ai/DeepSeek-R1-0528",
      "deepinfra.com/deepseek-ai/Janus-Pro-1B",
      "deepinfra.com/deepseek-ai/Janus-Pro-7B",
      "deepinfra.com/hexgrad/Kokoro-82M",
      "deepinfra.com/mistralai/Voxtral-Mini-3B-2507",
      "deepinfra.com/sesame/csm-1b"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://deepinfra.com/deepseek-ai/DeepSeek-R1-0528",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "deepinfra.com",
    "attribution": "DeepInfra official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-replicate",
    "name": "Replicate official pricing and model documentation",
    "match": [
      "replicate.com/black-forest-labs/flux-kontext-dev",
      "replicate.com/black-forest-labs/flux-krea-dev",
      "replicate.com/ibm-granite/granite-4.0-h-small",
      "replicate.com/ibm-granite/granite-4.1-8b",
      "replicate.com/ideogram-ai/ideogram-v2",
      "replicate.com/kwaivgi/kling-o1",
      "replicate.com/kwaivgi/kling-v2.1",
      "replicate.com/kwaivgi/kling-v2.1-master",
      "replicate.com/leonardoai/lucid-origin",
      "replicate.com/lightricks/ltx-2-distilled",
      "replicate.com/prunaai/p-video",
      "replicate.com/prunaai/z-image-turbo",
      "replicate.com/qwen/qwen-image",
      "replicate.com/qwen/qwen-image-2512",
      "replicate.com/resemble-ai/chatterbox",
      "replicate.com/stability-ai/stable-diffusion-3.5-large",
      "replicate.com/tencent/hunyuan-image-3",
      "replicate.com/wan-video/wan-2.1-1.3b",
      "replicate.com/wavespeedai/wan-2.1-t2v-720p",
      "replicate.com/prunaai/p-image",
      "replicate.com/prunaai/p-image-edit"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://replicate.com/ibm-granite/granite-4.0-h-small",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "replicate.com",
    "attribution": "Replicate official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-elevenlabs",
    "name": "ElevenLabs official pricing and model documentation",
    "match": [
      "elevenlabs.io/pricing/api",
      "elevenlabs.io/docs/overview/models",
      "elevenlabs.io/docs/changelog"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://elevenlabs.io/pricing/api",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "elevenlabs.io",
    "attribution": "ElevenLabs official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-cartesia",
    "name": "Cartesia official pricing and model documentation",
    "match": [
      "cartesia.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://cartesia.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "cartesia.ai",
    "attribution": "Cartesia official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-minimax",
    "name": "MiniMax official pricing and model documentation",
    "match": [
      "platform.minimax.io/docs/guides/pricing-paygo.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://platform.minimax.io/docs/guides/pricing-paygo.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "minimax.io",
    "attribution": "MiniMax official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-smallest",
    "name": "Smallest.ai official pricing and model documentation",
    "match": [
      "smallest.ai/pricing/models",
      "docs.smallest.ai/waves/model-cards/speech-to-text/pulse.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://smallest.ai/pricing/models",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "smallest.ai",
    "attribution": "Smallest.ai official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-hume",
    "name": "Hume AI official pricing and model documentation",
    "match": [
      "hume.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://hume.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "hume.ai",
    "attribution": "Hume AI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-azure",
    "name": "Microsoft Azure official pricing and model documentation",
    "match": [
      "azure.microsoft.com/en-us/pricing/details/ai-foundry-models/cohere",
      "azure.microsoft.com/en-us/pricing/details/ai-foundry-models/grok",
      "azure.microsoft.com/en-us/pricing/details/ai-foundry-models/microsoft",
      "azure.microsoft.com/en-us/pricing/details/speech",
      "learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://azure.microsoft.com/en-us/pricing/details/speech",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "azure.microsoft.com",
    "attribution": "Microsoft Azure official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-google-cloud",
    "name": "Google Cloud official pricing and model documentation",
    "match": [
      "cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing",
      "cloud.google.com/text-to-speech/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://cloud.google.com/text-to-speech/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "cloud.google.com",
    "attribution": "Google Cloud official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-boson",
    "name": "Boson AI official pricing and model documentation",
    "match": [
      "docs.boson.ai/pricing.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.boson.ai/pricing.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "boson.ai",
    "attribution": "Boson AI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-speechify",
    "name": "Speechify official pricing and model documentation",
    "match": [
      "speechify.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://speechify.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "speechify.ai",
    "attribution": "Speechify official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-breezeblue",
    "name": "BreezeBlue official pricing and model documentation",
    "match": [
      "breezeblue.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://breezeblue.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "breezeblue.ai",
    "attribution": "BreezeBlue official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-assemblyai",
    "name": "AssemblyAI official pricing and model documentation",
    "match": [
      "assemblyai.com/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://assemblyai.com/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "assemblyai.com",
    "attribution": "AssemblyAI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-gladia",
    "name": "Gladia official pricing and model documentation",
    "match": [
      "gladia.io/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://gladia.io/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "gladia.io",
    "attribution": "Gladia official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-soniox",
    "name": "Soniox official pricing and model documentation",
    "match": [
      "soniox.com/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://soniox.com/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "soniox.com",
    "attribution": "Soniox official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-speechmatics",
    "name": "Speechmatics official pricing and model documentation",
    "match": [
      "speechmatics.com/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://speechmatics.com/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "speechmatics.com",
    "attribution": "Speechmatics official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-deepgram",
    "name": "Deepgram official pricing and model documentation",
    "match": [
      "deepgram.com/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://deepgram.com/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "deepgram.com",
    "attribution": "Deepgram official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-mistral",
    "name": "Mistral AI official pricing and model documentation",
    "match": [
      "huggingface.co/mistralai/Mistral-Large-Instruct-2411/blob/ba78820945ae22361b0274cf0ae6d696c967c1a4/README.md",
      "mistral.ai/pricing/api",
      "arxiv.org/abs/2310.06825v1",
      "docs.mistral.ai/models/codestral-24-05",
      "docs.mistral.ai/models/mathstral-7b-0-1",
      "docs.mistral.ai/models/mistral-7b-0-3",
      "docs.mistral.ai/models/mixtral-8x22b-0-1-0-3",
      "docs.mistral.ai/models/mixtral-8x7b-0-1",
      "huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/1613bf01adb5e1c6fdc196b46e6b173eae75eb4a/README.md",
      "huggingface.co/mistralai/Magistral-Small-2506/blob/ad2fce5b4113139e1877dcadfd826c9262ad8e8c/README.md",
      "huggingface.co/mistralai/Ministral-8B-Instruct-2410/blob/2f494a194c5b980dfb9772cb92d26cbb671fce5a/README.md",
      "huggingface.co/mistralai/Mistral-7B-Instruct-v0.2/blob/63a8b081895390a26e140280378bc85ec8bce07a/README.md",
      "huggingface.co/mistralai/Mistral-Large-Instruct-2407/blob/a286006d554cb37a61d13c7ae61bc90cc1d372fc/README.md",
      "huggingface.co/mistralai/Mistral-Nemo-Instruct-2407/blob/04d8a90549d23fc6bd7f642064003592df51e9b3/README.md",
      "huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501/blob/9527884be6e5616bdd54de542f9ae13384489724/README.md",
      "huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503/blob/68faf511d618ef198fef186659617cfd2eb8e33a/README.md",
      "huggingface.co/mistralai/Mistral-Small-Instruct-2409/blob/4600506f6b13c7ef89e61a54263f4c9bf483de30/README.md",
      "docs.mistral.ai/models/model-cards/magistral-medium-1-2-25-09",
      "docs.mistral.ai/models/model-cards/magistral-medium-1-1-25-07",
      "docs.mistral.ai/models/model-cards/magistral-small-1-1-25-07",
      "docs.mistral.ai/models/mistral-large-4",
      "huggingface.co/mistralai/Mixtral-8x22B-v0.1/blob/e1cd34ff1747406fb2277635ed25242803009bc2/config.json"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://mistral.ai/pricing/api",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "mistral.ai",
    "attribution": "Mistral AI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-fal",
    "name": "fal.ai official pricing and model documentation",
    "match": [
      "fal.ai/models/bytedance/seedance-2.0/text-to-video",
      "fal.ai/models/fal-ai/bagel",
      "fal.ai/models/fal-ai/bytedance/seedance/v1.5/pro/text-to-video",
      "fal.ai/models/fal-ai/bytedance/seedance/v1/pro/fast/text-to-video",
      "fal.ai/models/fal-ai/cohere-transcribe",
      "fal.ai/models/fal-ai/ernie-image",
      "fal.ai/models/fal-ai/ernie-image/turbo",
      "fal.ai/models/fal-ai/gemini-3.1-flash-tts",
      "fal.ai/models/fal-ai/hidream-i1-full",
      "fal.ai/models/fal-ai/hidream-o1-image",
      "fal.ai/models/fal-ai/hunyuan-image/v2.1/text-to-image",
      "fal.ai/models/fal-ai/hunyuan-image/v3/instruct/text-to-image",
      "fal.ai/models/fal-ai/hunyuan-video-v1.5/image-to-video",
      "fal.ai/models/fal-ai/hunyuan-video-v1.5/text-to-video",
      "fal.ai/models/fal-ai/inworld-tts",
      "fal.ai/models/fal-ai/kandinsky5-pro/text-to-video",
      "fal.ai/models/fal-ai/kling-video/o3/pro/text-to-video",
      "fal.ai/models/fal-ai/kling-video/v2.5-turbo/pro/text-to-video",
      "fal.ai/models/fal-ai/kling-video/v2.6/pro/text-to-video",
      "fal.ai/models/fal-ai/kling-video/v3/pro/text-to-video",
      "fal.ai/models/fal-ai/longcat-image",
      "fal.ai/models/fal-ai/longcat-image/edit",
      "fal.ai/models/fal-ai/longcat-video/text-to-video/720p",
      "fal.ai/models/fal-ai/luma-photon",
      "fal.ai/models/fal-ai/lumina-image/v2",
      "fal.ai/models/fal-ai/minimax/video-01/image-to-video",
      "fal.ai/models/fal-ai/omnigen-v2",
      "fal.ai/models/fal-ai/pika/v2.2/text-to-video",
      "fal.ai/models/fal-ai/qwen-image-edit",
      "fal.ai/models/fal-ai/qwen-image-edit-2509",
      "fal.ai/models/fal-ai/qwen-image-edit-2511",
      "fal.ai/models/fal-ai/stable-diffusion-v3-medium",
      "fal.ai/models/fal-ai/vidu/q2/image-to-video/turbo",
      "fal.ai/models/fal-ai/wan/v2.2-a14b/image-to-video",
      "fal.ai/models/fal-ai/wan/v2.2-a14b/text-to-video",
      "fal.ai/models/fal-ai/z-image/base",
      "fal.ai/models/openai/gpt-image-2.5/flare/text-to-image",
      "fal.ai/models/openai/gpt-image-2.5/sunburst/text-to-image",
      "fal.ai/models/reve/2.0/text-to-image",
      "fal.ai/models/reve/2.1/text-to-image"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://fal.ai/models/bytedance/seedance-2.0/text-to-video",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "fal.ai",
    "attribution": "fal.ai official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-kling",
    "name": "Kling AI official pricing and model documentation",
    "match": [
      "kling.ai/dev/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://kling.ai/dev/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "kling.ai",
    "attribution": "Kling AI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-vidu",
    "name": "Vidu (ShengShu) official pricing and model documentation",
    "match": [
      "platform.vidu.com/docs/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://platform.vidu.com/docs/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "vidu.com",
    "attribution": "Vidu (ShengShu) official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-pixverse",
    "name": "PixVerse official pricing and model documentation",
    "match": [
      "docs.platform.pixverse.ai/model-pricing-796039m0"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.platform.pixverse.ai/model-pricing-796039m0",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "pixverse.ai",
    "attribution": "PixVerse official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-ltx",
    "name": "Lightricks LTX official pricing and model documentation",
    "match": [
      "docs.ltx.io/pricing.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.ltx.io/pricing.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "ltx.io",
    "attribution": "Lightricks LTX official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-bfl",
    "name": "Black Forest Labs official pricing and model documentation",
    "match": [
      "bfl.ai/pricing",
      "docs.bfl.ai/quick_start/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.bfl.ai/quick_start/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "bfl.ai",
    "attribution": "Black Forest Labs official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-runway",
    "name": "Runway official pricing and model documentation",
    "match": [
      "docs.dev.runwayml.com/guides/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.dev.runwayml.com/guides/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "runwayml.com",
    "attribution": "Runway official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-recraft",
    "name": "Recraft official pricing and model documentation",
    "match": [
      "recraft.ai/docs/api-reference/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://www.recraft.ai/docs/api-reference/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "recraft.ai",
    "attribution": "Recraft official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-zai",
    "name": "Z.ai official pricing and model documentation",
    "match": [
      "docs.z.ai/guides/overview/pricing",
      "docs.z.ai/guides/vlm/glm-5v-turbo.md",
      "huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/README.md",
      "huggingface.co/zai-org/GLM-4.6V/blob/4e2d47eb0b41c5280d8294b17cef9e94fdcfff46/README.md",
      "huggingface.co/zai-org/glm-4-9b-chat/blob/bd8234fe5e0c09c48637a92abb0c797cb5fa0e73/README.md",
      "huggingface.co/zai-org/chatglm3-6b-128k/blob/6381a643d74e3e6a1763580bb643e488dd278b98/config.json",
      "huggingface.co/zai-org/glm-4-9b-chat-1m/blob/f8579ac096986e44ba24a5060d2fb79bedb56ef3/config.json"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.z.ai/guides/overview/pricing",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "z.ai",
    "attribution": "Z.ai official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-byteplus",
    "name": "BytePlus ModelArk official pricing and model documentation",
    "match": [
      "docs.byteplus.com/en/docs/ModelArk/1544106"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.byteplus.com/en/docs/ModelArk/1544106",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "byteplus.com",
    "attribution": "BytePlus ModelArk official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-lg-ai-research",
    "name": "LG AI Research official pricing and model documentation",
    "match": [
      "huggingface.co/LGAI-EXAONE/EXAONE-4.0-1.2B/blob/3abf2810673c7c0778df64a73c2d52eab32d91c4/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B/blob/a1d54d1c148c30881ed27e035b650da489b51b92/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-4.5-33B/blob/570aa4b15a4f45ba1133072b45f50198f6e3b4fd/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct/blob/ccce25bd39c141fe053e0bc75818a8f5fe962802/README.md",
      "huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B/blob/61e6d578eb102b578e5704e2916ac841df9eca0a/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct/blob/553ea250b9a5317231459279d5847d6cf955b9aa/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-3.5-32B-Instruct/blob/3d27aab477c042df2e174d36eea47b77c8871ef0/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-Deep-2.4B/blob/69d80cdf4784e2c182854d83e9b2d02494626a3e/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-Deep-7.8B/blob/8a120673fc7309b55be2175a2e61f9a0839e4f61/README.md",
      "huggingface.co/LGAI-EXAONE/EXAONE-Deep-32B/blob/4e9a4c20b6c3f9093777bfe6a98b2be508abb919/README.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B/blob/a1d54d1c148c30881ed27e035b650da489b51b92/README.md",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "lgresearch.ai",
    "attribution": "LG AI Research official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-liquid-ai",
    "name": "Liquid AI official pricing and model documentation",
    "match": [
      "huggingface.co/LiquidAI/LFM2-24B-A2B/blob/a3bbacd91a678b97712f0e323e52f8c24ba29542/README.md",
      "huggingface.co/LiquidAI/LFM2.5-2.6B/blob/654f9463ce32b05d0429d76fe1f580b27d4c1ac0/README.md",
      "huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct/blob/0f604ada3f766f9f257460c4c9f0b5d6f69d431b/README.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://huggingface.co/LiquidAI/LFM2-24B-A2B/blob/a3bbacd91a678b97712f0e323e52f8c24ba29542/README.md",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "liquid.ai",
    "attribution": "Liquid AI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-baidu",
    "name": "Baidu official pricing and model documentation",
    "match": [
      "huggingface.co/baidu/ERNIE-4.5-300B-A47B-PT/blob/47bbaf1c9e356313c3cb2ded9078a494fb152f0e/README.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-PT/blob/47bbaf1c9e356313c3cb2ded9078a494fb152f0e/README.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "baidu.com",
    "attribution": "Baidu official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-tii",
    "name": "Technology Innovation Institute official pricing and model documentation",
    "match": [
      "huggingface.co/tiiuae/Falcon-H1R-7B/blob/a6f74bf181389908efd6970878d7ee2b42f5d417/config.json",
      "huggingface.co/tiiuae/Falcon3-10B-Instruct/blob/8799bc6aec0152757221dc6b272d824642db6202/README.md",
      "huggingface.co/tiiuae/Falcon3-1B-Instruct/blob/28ba2251970a01dd1edc7ba7dad2eb71216ccfdf/README.md",
      "huggingface.co/tiiuae/Falcon3-3B-Instruct/blob/411bb94318f94f7a5735b77109f456b1e74b42a1/README.md",
      "huggingface.co/tiiuae/Falcon3-7B-Instruct/blob/1e57a0ecd176c7c139f289c60a74e57f887c3dfb/README.md",
      "huggingface.co/tiiuae/falcon-180B-chat/blob/6882299beda36df40e683c4478edfcc94f2cde5f/README.md",
      "huggingface.co/tiiuae/falcon-40b/blob/05ab2ee8d6b593bdbab17d728de5c028a7a94d83/README.md",
      "huggingface.co/tiiuae/falcon-7b/blob/ec89142b67d748a1865ea4451372db8313ada0d8/README.md",
      "huggingface.co/tiiuae/falcon-rw-1b/blob/e4b9872bb803165eb22f0a867d4e6a64d34fce19/README.md",
      "huggingface.co/tiiuae/Falcon-H1-1.5B-Instruct/blob/80ebc50d7799a440b96c93bb6686a3924a09b0cb/config.json",
      "huggingface.co/tiiuae/Falcon-H1-34B-Instruct/blob/6e6890e26cba1f58b1e6ee70d654ddf054f9fce8/config.json",
      "huggingface.co/tiiuae/Falcon-H1-3B-Instruct/blob/01087ec4c132d7f186908716b3530ea187f062a1/config.json",
      "huggingface.co/tiiuae/Falcon-H1-7B-Instruct/blob/41e72f27effbab80cd45b6e884688452253a3686/config.json",
      "huggingface.co/tiiuae/falcon-11B/blob/a4ab868d7e8d53844009657bf7e755933ccccf8f/config.json"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://huggingface.co/tiiuae/Falcon-H1R-7B/blob/a6f74bf181389908efd6970878d7ee2b42f5d417/config.json",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "tii.ae",
    "attribution": "Technology Innovation Institute official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "id": "speech-qwen-audio3-tts-report",
    "name": "Qwen-Audio-3.0-TTS technical report",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2607.23938v1",
    "match": [
      "arxiv.org/html/2607.23938v1"
    ],
    "icon": "arxiv.org",
    "license_url": "/licenses/speech-qwen-audio3-tts-report.txt",
    "attribution": "Bajian Xiang et al. (Alibaba Token Foundry); Qwen-Audio-3.0-TTS: Freely Controllable and Highly Robust Speech Synthesis with Multi-Stage Training Paradigm; arXiv:2607.23938v1. CC BY 4.0. Changes: selected own-run CV3-Eval and SEED-TTS-Eval cells, percent/100, derived equal-language means (marked). No endorsement.",
    "quote": "arXiv abstract page for 2607.23938v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' (CC BY 4.0) for this version; the HTML header states 'License: CC BY 4.0'. Applies to the report's text and tables; not a claim about model weights. The report's quoted Artificial Analysis Elo is excluded.",
    "scope": "Extended 2026-10-09: Table 3 ERes2Net SIM (in parentheses) for the authors' own model and the cells marked † ('ERes2Net SIM scores were computed by us using the publicly released models'); unmarked parenthesised values not used. Pass 8 (2026-10-09): adds the LongCat-AudioDiT-3.5B row (Table 3 † ERes2Net SIM, Table 5 hard-zh/hard-en error) and Table 5 hard-en ERes2Net SIM and DNSMOS for all catalogued systems (bench ids cv3-eval-hard-en-eres2net-sim, cv3-eval-hard-en-dnsmos). No new URL; match unchanged."
  },
  {
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "id": "speech-zonos2-report-bysa",
    "name": "ZONOS2 Technical Report (ZTTS1-Eval results)",
    "kind": "paper",
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2606.24320v2",
    "match": [
      "arxiv.org/html/2606.24320v2"
    ],
    "icon": "arxiv.org",
    "license_url": "/licenses/speech-zonos2-report-bysa.txt",
    "attribution": "Gabriel Clark, Sofian Mejjoute, Mohamed Osman, George Close, Beren Millidge (Zyphra); ZONOS2 Technical Report; arXiv:2606.24320v2. CC BY-SA 4.0. Changes: selected Table III/IV/VII cells, percent/100, derived equal-language means over stated common languages, UTMOS normalized (x-1)/4. This numerical adaptation is licensed CC BY-SA 4.0. No endorsement.",
    "quote": "arXiv abstract page for 2606.24320v2 links 'http://creativecommons.org/licenses/by-sa/4.0/'; HTML header 'License: CC BY-SA 4.0'. ShareAlike: adapted numerical material must be shared under CC BY-SA 4.0 (compatible with DATA-LICENSE.md CC BY-SA 4.0)."
  },
  {
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "id": "speech-rw-voice-eq-report",
    "name": "RW-Voice-EQ Bench (Hume AI)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2607.14846v1",
    "match": [
      "arxiv.org/html/2607.14846v1",
      "arxiv.org/html/2607.14846v1/TTS_Top5_Membership_Matrix.svg",
      "arxiv.org/html/2607.14846v1/avg_wer_bar_top10.svg"
    ],
    "icon": "arxiv.org",
    "license_url": "/licenses/speech-rw-voice-eq-report.txt",
    "attribution": "David Ayllon, Alice Baird, Jeffrey Brooks et al. (Hume AI Research); RW-Voice-EQ Bench: A Real World Benchmark for Evaluating Voice AI Systems; arXiv:2607.14846v1. CC BY 4.0. Changes: exact printed values transcribed from Figure 5 and Figure 8 (paper figures), rating normalized (x-1)/4, percent/100, model identities mapped via Appendix Tables 4 and 7. No endorsement.",
    "quote": "arXiv abstract page for 2607.14846v1 links 'http://creativecommons.org/licenses/by/4.0/'; HTML header 'License: CC BY 4.0'. The figures are part of the licensed paper; only printed labels were transcribed (no bar-height estimation). The linked HF Space/private dataset is not covered and was not used."
  },
  {
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "id": "speech-higgs3-sglang-apache",
    "name": "Higgs Audio v3 comparison tables in SGLang-Omni cookbook (Boson AI contribution)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/sgl-project/sglang-omni/blob/0aafd76b4ef488f1c01987cc856cea77af129d61/LICENSE",
    "match": [
      "github.com/sgl-project/sglang-omni/blob/0aafd76b4ef488f1c01987cc856cea77af129d61/docs/cookbook/higgs_tts.md"
    ],
    "icon": "huggingface.co",
    "license_url": "/licenses/speech-higgs3-sglang-apache.txt",
    "attribution": "Boson AI (contributed by Silin Meng, PR #684) in sgl-project/sglang-omni, docs/cookbook/higgs_tts.md at 0aafd76b4ef488f1c01987cc856cea77af129d61. Apache License 2.0. Changes: selected macro WER/CER cells for catalogued models, x100 values divided by 100. Retain LICENSE. No endorsement.",
    "quote": "Repository root LICENSE at this commit is the Apache License 2.0 and covers the documentation file; the file was committed by a Boson AI employee (GitHub profile company 'Boson AI') as the model-card comparison tables, i.e. the rights holder's own contribution under Apache-2.0 section 5. The separate HF model card (NC licence) is used only to corroborate that the numbers match the released checkpoint."
  },
  {
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "id": "speech-lemas-report-bysa",
    "name": "LEMAS multilingual audio suite report",
    "kind": "paper",
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2601.04233v1",
    "match": [
      "arxiv.org/html/2601.04233v1"
    ],
    "icon": "arxiv.org",
    "license_url": "/licenses/speech-lemas-report-bysa.txt",
    "attribution": "Zhiyuan Zhao et al.; LEMAS: A 150K-Hour Large-scale Extensible Multilingual Audio Suite with Generative Speech Models; arXiv:2601.04233v1. CC BY-SA 4.0. Changes: selected the OpenAudio-S1-Mini average WER from Table 4, percent/100. This numerical adaptation is licensed CC BY-SA 4.0. No endorsement.",
    "quote": "arXiv abstract page for 2601.04233v1 links 'http://creativecommons.org/licenses/by-sa/4.0/'. ShareAlike applies to the adapted numerical material."
  },
  {
    "id": "speech-emergent-v1-paper-ccby",
    "name": "EmergentTTS-Eval v1 paper (arXiv:2505.23009v1) · Table 1 automatic MOS",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2505.23009v1",
    "checked": "2026-09-25",
    "attribution": "Ruskin Raj Manku et al. (Boson AI); EmergentTTS-Eval: Evaluating TTS Models on Complex Prosodic, Expressiveness, and Linguistic Challenges Using Model-as-a-Judge; arXiv:2505.23009v1. CC BY 4.0. Table 1 MOS column transcribed; raw WV-MOS retained and normalized (MOS-1)/4. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2505.23009v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "Versioned arXiv abstract page (only version v1, submitted 29 May 2025) declares the article licence via 'view license' -> http://creativecommons.org/licenses/by/4.0/; HTML rendering states 'License: CC BY 4.0'. Covers the report including its numerical tables. Not a claim about model or audio licences.",
    "license_url": "/licenses/speech-emergent-v1-paper-ccby.txt"
  },
  {
    "id": "speech-ztts1-eval-apache",
    "name": "ZTTS1-Eval results snapshot (Zyphra)",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Zyphra/ZTTS1-Eval/blob/93660731c7b8d5e70935975fe7a6c7ed48451a7b/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright 2025 Zyphra. ZTTS1-Eval, Apache License 2.0; retain LICENSE and NOTICE. Selected FLEURS-R zero-shot English TTSDS2 PROSODY means, divided by 100. VoxBlink2 (non-commercial data) results not used. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/Zyphra/ZTTS1-Eval/blob/93660731c7b8d5e70935975fe7a6c7ed48451a7b/RESULTS_CURRENT.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Root LICENSE is the Apache License, Version 2.0. DATA_LICENSE.md: 'The code in this repository is licensed under Apache-2.0 (see LICENSE). The audio and text data under data/ is not covered by that license' — the only carve-out is data/; RESULTS_CURRENT.md sits at the repository root. Only the FLEURS-R (CC-BY-4.0 source data) English table is used; VoxBlink2 (CC-BY-NC-SA 4.0 source data) rows are excluded.",
    "license_url": "/licenses/speech-ztts1-eval-apache.txt"
  },
  {
    "id": "speech-stt-granite5-report-bysa",
    "name": "Design of the IBM Granite 5.0 TurboCTC ASR Model (arXiv:2609.20104v1)",
    "kind": "paper",
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2609.20104v1",
    "checked": "2026-09-25",
    "attribution": "Brian Kingsbury, George Saon, Masayuki Suzuki, Hong-Kwang J. Kuo, Takashi Fukuda, Samuel Thomas, Vishal Sunder, Avihu Dekel et al. (IBM); arXiv:2609.20104v1, CC BY-SA 4.0. Table 3 final-stage AVG WER transcribed, percent divided by 100. Share-alike: adapted numerical record remains CC BY-SA 4.0. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2609.20104v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2609.20104v1 declares 'License: CC BY-SA 4.0' (http://creativecommons.org/licenses/by-sa/4.0/) for the versioned report, including its numerical tables. First-party (lab) measurement of IBM's own model. Not a claim about Open ASR leaderboard data, charts or weights.",
    "license_url": "/licenses/speech-stt-granite5-report-bysa.txt"
  },
  {
    "id": "speech-stt-qwen-audio3-asr-report",
    "name": "Qwen-Audio-3.0-ASR Technical Report (arXiv:2609.07549v2)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2609.07549v2",
    "checked": "2026-09-25",
    "attribution": "Qwen-Audio ASR Team (Alibaba Token Foundry); Qwen-Audio-3.0-ASR Technical Report; arXiv:2609.07549v2. CC BY 4.0. Table 1 Panel B (own unified API evaluation) transcribed; equal-set means derived; percent divided by 100. Panel A (copied official results) and Table 2 (provenance not stated) excluded. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2609.07549v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2609.07549v2 (and v1) declares 'License: CC BY 4.0' (http://creativecommons.org/licenses/by/4.0/). Only Panel B, which the paper states was 'obtained using our unified API-based evaluation pipeline on the same test sets', is used; Panel A rows 'taken from official public repositories or technical reports' are copied and excluded.",
    "license_url": "/licenses/speech-stt-qwen-audio3-asr-report.txt"
  },
  {
    "id": "speech-stt-verli-mit",
    "name": "verliapp/stt-benchmark · independent STT accuracy benchmark",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/verliapp/stt-benchmark/blob/bfb51427ba12380ff1e3620a8333c10f18fa4280/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2026 Inkvi (verliapp/stt-benchmark). MIT. Selected corpus WERs from committed results/results_*.json, percent divided by 100; retain MIT notice. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/verliapp/stt-benchmark/blob/bfb51427ba12380ff1e3620a8333c10f18fa4280/results/results_librispeech.json",
      "github.com/verliapp/stt-benchmark/blob/bfb51427ba12380ff1e3620a8333c10f18fa4280/results/results_earnings22.json",
      "github.com/verliapp/stt-benchmark/blob/bfb51427ba12380ff1e3620a8333c10f18fa4280/results/results_spgispeech.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Root LICENSE: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies'; README: 'MIT licensed.' Committed result JSON files are repository files under this grant. No dataset audio or transcripts reused.",
    "license_url": "/licenses/speech-stt-verli-mit.txt"
  },
  {
    "id": "speech-stt-koedesk-ccby",
    "name": "koedesk STT Bench v1 (Track A, FLEURS first-200)",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://huggingface.co/datasets/koedesk/stt-bench/blob/c97de119c0511900d559bde5024c2fdc31be0b2c/README.md",
    "checked": "2026-09-25",
    "attribution": "koedesk STT Bench, (c) 2026 Guide Inc. (koedesk), CC BY 4.0 (https://huggingface.co/datasets/koedesk/stt-bench). Reference texts derive from FLEURS (CC BY 4.0). Changes: selected engine rows, derived an equal-language mean over nine languages covered by every selected engine, error rates kept as fractions. Conflict of interest disclosed by source (koedesk uses ElevenLabs Scribe v2). No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/datasets/koedesk/stt-bench/blob/c97de119c0511900d559bde5024c2fdc31be0b2c/scores.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Dataset card metadata 'license: cc-by-4.0' and License section: 'This dataset (transcriptions, scores, documentation): CC BY 4.0, (c) 2026 Guide Inc. (koedesk).' scores.json is included verbatim in the dataset (identical SHA-256 to GitHub results/scores.json, whose repository README says 'Code: MIT. Benchmark documents: CC-BY 4.0.').",
    "license_url": "/licenses/speech-stt-koedesk-ccby.txt"
  },
  {
    "id": "speech-stt-raven-mit",
    "name": "raven.eval · German ASR WER (public datasets)",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/PhilflowIO/raven.eval/blob/f9f621e23b0e83e29aa18e8ffcd4656805c68145/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2026 Philflow / raven.eval. MIT. Four flozi-strict corpus WERs for xai/grok-voice-transcribe-2.0 from BENCHMARKS.md (re-scored artifacts), equal-set mean derived, percent divided by 100. Retain MIT notice. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/PhilflowIO/raven.eval/blob/f9f621e23b0e83e29aa18e8ffcd4656805c68145/BENCHMARKS.md"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Root LICENSE (MIT): 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute'. BENCHMARKS.md and artifacts/*/expected.json are repository files merged to master (not issue/PR comments).",
    "license_url": "/licenses/speech-stt-raven-mit.txt"
  },
  {
    "id": "speech-stt-bangla-riverborn-mit",
    "name": "Bangla Speech-to-Text Benchmark v1.0 (Riverborn)",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/riverbornai/bangla-speech-to-text-benchmark/blob/cc73eb073b065bc4c7301672a2013c70d446450c/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Bangla Speech-to-Text Benchmark v1.0 by Riverborn Limited (riverbornai/bangla-speech-to-text-benchmark), MIT. Test set: BanSpeech (SUST CSE Speech). Selected batch corpus CERs from results/leaderboard.csv; no reference transcripts reused. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/riverbornai/bangla-speech-to-text-benchmark/blob/cc73eb073b065bc4c7301672a2013c70d446450c/results/leaderboard.csv"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Root LICENSE is MIT ('Permission is hereby granted, free of charge ... to deal in the Software without restriction'); README License section: 'Code: MIT'; the only carve-out is 'Test-set transcripts (data/test_set.csv and the ref columns in results/*per_file.csv): from BanSpeech, redistributed under CC BY-NC 4.0' and audio not redistributed. leaderboard.csv holds only numerical scores and is under the root MIT grant; no ref text is reused.",
    "license_url": "/licenses/speech-stt-bangla-riverborn-mit.txt"
  },
  {
    "id": "speech-stt-perle-cs-report",
    "name": "Benchmarking Commercial ASR Systems on Code-Switching Speech (arXiv:2605.19069v3)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2605.19069v3",
    "checked": "2026-09-25",
    "attribution": "Sajjad Abdoli, Ghassan Al-Sumaidaee, Clayton W. Taylor, Ahmad ElShiekh, Ahmed Rashad (Perle AI); arXiv:2605.19069v3, CC BY 4.0. Table 4 mean WER transcribed, percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/pdf/2605.19069v3"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2605.19069v3 declares 'License: CC BY 4.0' (http://creativecommons.org/licenses/by/4.0/). The authors ran all five commercial APIs themselves in May 2026 (Table 9 lists exact parameters). No arXiv HTML rendering exists for v3, so the versioned PDF is the pinned locator.",
    "license_url": "/licenses/speech-stt-perle-cs-report.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-14b-fp8-dynamic",
    "name": "RedHatAI/Qwen2.5-14B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-14B-FP8-dynamic/blob/99a1817b207d2d5684d745e30b3e06473cea1d10/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-14B-FP8-dynamic model card, revision 99a1817b207d2d5684d745e30b3e06473cea1d10. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-14B-FP8-dynamic/blob/99a1817b207d2d5684d745e30b3e06473cea1d10/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-14b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-14b-instruct-w8a8",
    "name": "RedHatAI/Qwen2.5-14B-Instruct-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-14B-Instruct-quantized.w8a8/blob/a9d77c2a8d0706cebcc7a5dcdd8221c2bda0ca6d/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-14B-Instruct-quantized.w8a8 model card, revision a9d77c2a8d0706cebcc7a5dcdd8221c2bda0ca6d. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-14B-Instruct-quantized.w8a8/blob/a9d77c2a8d0706cebcc7a5dcdd8221c2bda0ca6d/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-14b-instruct-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-14b-w8a8",
    "name": "RedHatAI/Qwen2.5-14B-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-14B-quantized.w8a8/blob/b820dc00e4ebd4da94d17d3087a379b2bba40cef/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-14B-quantized.w8a8 model card, revision b820dc00e4ebd4da94d17d3087a379b2bba40cef. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-14B-quantized.w8a8/blob/b820dc00e4ebd4da94d17d3087a379b2bba40cef/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-14b-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-32b-w8a16",
    "name": "RedHatAI/Qwen2.5-32B-quantized.w8a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-32B-quantized.w8a16/blob/718f3680200a636693a0858d145d8651b2264c52/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-32B-quantized.w8a16 model card, revision 718f3680200a636693a0858d145d8651b2264c52. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-32B-quantized.w8a16/blob/718f3680200a636693a0858d145d8651b2264c52/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 9: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-32b-w8a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-7b-fp8-dynamic",
    "name": "RedHatAI/Qwen2.5-7B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-7B-FP8-dynamic/blob/972b5206a1ad2891762dbe19e89651f9a7c536bd/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-7B-FP8-dynamic model card, revision 972b5206a1ad2891762dbe19e89651f9a7c536bd. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-7B-FP8-dynamic/blob/972b5206a1ad2891762dbe19e89651f9a7c536bd/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-7b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-7b-instruct-fp8-dynamic",
    "name": "RedHatAI/Qwen2.5-7B-Instruct-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-7B-Instruct-FP8-dynamic/blob/a4f1d5442ea284bc8e5a3b3e1a4d528f7d6caacb/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-7B-Instruct-FP8-dynamic model card, revision a4f1d5442ea284bc8e5a3b3e1a4d528f7d6caacb. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-7B-Instruct-FP8-dynamic/blob/a4f1d5442ea284bc8e5a3b3e1a4d528f7d6caacb/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 47: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-7b-instruct-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-7b-instruct-w4a16",
    "name": "RedHatAI/Qwen2.5-7B-Instruct-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-7B-Instruct-quantized.w4a16/blob/92e73453c88d10975a06b42789f63933a6c3f449/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-7B-Instruct-quantized.w4a16 model card, revision 92e73453c88d10975a06b42789f63933a6c3f449. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-7B-Instruct-quantized.w4a16/blob/92e73453c88d10975a06b42789f63933a6c3f449/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 48: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-7b-instruct-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-7b-instruct-w8a8",
    "name": "RedHatAI/Qwen2.5-7B-Instruct-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-7B-Instruct-quantized.w8a8/blob/dae5e0d2d3cc641da65943fa6ef02e5e02fd4d64/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-7B-Instruct-quantized.w8a8 model card, revision dae5e0d2d3cc641da65943fa6ef02e5e02fd4d64. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-7B-Instruct-quantized.w8a8/blob/dae5e0d2d3cc641da65943fa6ef02e5e02fd4d64/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 48: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-7b-instruct-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-7b-w4a16",
    "name": "RedHatAI/Qwen2.5-7B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-7B-quantized.w4a16/blob/88338db35a8b86592d201e91aac04e5d756500ab/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-7B-quantized.w4a16 model card, revision 88338db35a8b86592d201e91aac04e5d756500ab. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-7B-quantized.w4a16/blob/88338db35a8b86592d201e91aac04e5d756500ab/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-7b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-7b-w8a16",
    "name": "RedHatAI/Qwen2.5-7B-quantized.w8a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-7B-quantized.w8a16/blob/0553881e16f1700d3d007227b986661497e8e7bc/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-7B-quantized.w8a16 model card, revision 0553881e16f1700d3d007227b986661497e8e7bc. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-7B-quantized.w8a16/blob/0553881e16f1700d3d007227b986661497e8e7bc/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 9: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-7b-w8a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen2-5-7b-w8a8",
    "name": "RedHatAI/Qwen2.5-7B-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen2.5-7B-quantized.w8a8/blob/5c84ad661e3c4f382aa885cfc32e57c7a78ae915/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen2.5-7B-quantized.w8a8 model card, revision 5c84ad661e3c4f382aa885cfc32e57c7a78ae915. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen2.5-7B-quantized.w8a8/blob/5c84ad661e3c4f382aa885cfc32e57c7a78ae915/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen2-5-7b-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-0-6b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-0.6B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-0.6B-FP8-dynamic/blob/068a9040b238d65f5c1064f10232bb15b96c0ff0/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-0.6B-FP8-dynamic model card, revision 068a9040b238d65f5c1064f10232bb15b96c0ff0. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-0.6B-FP8-dynamic/blob/068a9040b238d65f5c1064f10232bb15b96c0ff0/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-0-6b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-0-6b-w4a16",
    "name": "RedHatAI/Qwen3-0.6B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-0.6B-quantized.w4a16/blob/3bceedd23534ac8ef59952f72c94790b27938e42/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-0.6B-quantized.w4a16 model card, revision 3bceedd23534ac8ef59952f72c94790b27938e42. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-0.6B-quantized.w4a16/blob/3bceedd23534ac8ef59952f72c94790b27938e42/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-0-6b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-1-7b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-1.7B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-1.7B-FP8-dynamic/blob/e776dcc126bb25d3d5af2b619acaad8a94332a86/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-1.7B-FP8-dynamic model card, revision e776dcc126bb25d3d5af2b619acaad8a94332a86. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-1.7B-FP8-dynamic/blob/e776dcc126bb25d3d5af2b619acaad8a94332a86/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-1-7b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-1-7b-w4a16",
    "name": "RedHatAI/Qwen3-1.7B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-1.7B-quantized.w4a16/blob/f8467681051842a5be10a8478d8cfdc5facb0af3/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-1.7B-quantized.w4a16 model card, revision f8467681051842a5be10a8478d8cfdc5facb0af3. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-1.7B-quantized.w4a16/blob/f8467681051842a5be10a8478d8cfdc5facb0af3/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-1-7b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-14b-fp8-block",
    "name": "RedHatAI/Qwen3-14B-FP8-block model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-14B-FP8-block/blob/dc7381d4ba723321bf2db13ac62fc2db50cc7815/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-14B-FP8-block model card, revision dc7381d4ba723321bf2db13ac62fc2db50cc7815. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-14B-FP8-block/blob/dc7381d4ba723321bf2db13ac62fc2db50cc7815/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-14b-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-14b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-14B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-14B-FP8-dynamic/blob/8bae2a5f4cf23071026e82c6625c0f93dde165f9/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-14B-FP8-dynamic model card, revision 8bae2a5f4cf23071026e82c6625c0f93dde165f9. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-14B-FP8-dynamic/blob/8bae2a5f4cf23071026e82c6625c0f93dde165f9/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-14b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-14b-nvfp4",
    "name": "RedHatAI/Qwen3-14B-NVFP4 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-14B-NVFP4/blob/cc120e7bc68557313002275d51db65096ca24303/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-14B-NVFP4 model card, revision cc120e7bc68557313002275d51db65096ca24303. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-14B-NVFP4/blob/cc120e7bc68557313002275d51db65096ca24303/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-14b-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-14b-w4a16",
    "name": "RedHatAI/Qwen3-14B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-14B-quantized.w4a16/blob/98ce4ba52bc02e37e2beb0b0a5f11d7f0e589685/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-14B-quantized.w4a16 model card, revision 98ce4ba52bc02e37e2beb0b0a5f11d7f0e589685. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-14B-quantized.w4a16/blob/98ce4ba52bc02e37e2beb0b0a5f11d7f0e589685/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-14b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-fp8-block",
    "name": "RedHatAI/Qwen3-30B-A3B-FP8-block model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-FP8-block/blob/6b2399be90eff484848cbbabd684930e01400d54/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-FP8-block model card, revision 6b2399be90eff484848cbbabd684930e01400d54. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-FP8-block/blob/6b2399be90eff484848cbbabd684930e01400d54/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-30B-A3B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-FP8-dynamic/blob/10a439027a4fd228adc03a7bffb1aac10a179938/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-FP8-dynamic model card, revision 10a439027a4fd228adc03a7bffb1aac10a179938. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-FP8-dynamic/blob/10a439027a4fd228adc03a7bffb1aac10a179938/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-instruct-2507-w4a16",
    "name": "RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w4a16/blob/e9c59cdcca9d73f24adeda09660af904a79f509f/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w4a16 model card, revision e9c59cdcca9d73f24adeda09660af904a79f509f. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w4a16/blob/e9c59cdcca9d73f24adeda09660af904a79f509f/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-instruct-2507-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-instruct-2507-w8a8",
    "name": "RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w8a8/blob/b2b37fb1c4cd5a5769b841ef1f672627b8304d8e/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w8a8 model card, revision b2b37fb1c4cd5a5769b841ef1f672627b8304d8e. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-Instruct-2507-quantized.w8a8/blob/b2b37fb1c4cd5a5769b841ef1f672627b8304d8e/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-instruct-2507-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-nvfp4",
    "name": "RedHatAI/Qwen3-30B-A3B-NVFP4 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-NVFP4/blob/a4df9f017835830cbb588cfe72bc4e50294183c0/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-NVFP4 model card, revision a4df9f017835830cbb588cfe72bc4e50294183c0. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-NVFP4/blob/a4df9f017835830cbb588cfe72bc4e50294183c0/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-thinking-2507-w4a16",
    "name": "RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w4a16/blob/98ffbaed9192e305240ff9fb6647ac1ea0abb467/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w4a16 model card, revision 98ffbaed9192e305240ff9fb6647ac1ea0abb467. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w4a16/blob/98ffbaed9192e305240ff9fb6647ac1ea0abb467/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-thinking-2507-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-thinking-2507-w8a8",
    "name": "RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w8a8/blob/fb485fe24f99380b6f480ac767f671cd5c740908/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w8a8 model card, revision fb485fe24f99380b6f480ac767f671cd5c740908. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-Thinking-2507-quantized.w8a8/blob/fb485fe24f99380b6f480ac767f671cd5c740908/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-thinking-2507-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-30b-a3b-w4a16",
    "name": "RedHatAI/Qwen3-30B-A3B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-30B-A3B-quantized.w4a16/blob/1c842a5000febac4b335c94a04789a9d4756a18e/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-30B-A3B-quantized.w4a16 model card, revision 1c842a5000febac4b335c94a04789a9d4756a18e. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-30B-A3B-quantized.w4a16/blob/1c842a5000febac4b335c94a04789a9d4756a18e/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-30b-a3b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-32b-fp8-block",
    "name": "RedHatAI/Qwen3-32B-FP8-block model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-32B-FP8-block/blob/90c5558f4bc66d558f691d0b3077d39b1dd70dec/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-32B-FP8-block model card, revision 90c5558f4bc66d558f691d0b3077d39b1dd70dec. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-32B-FP8-block/blob/90c5558f4bc66d558f691d0b3077d39b1dd70dec/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-32b-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-32b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-32B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-32B-FP8-dynamic/blob/c6732fc26128341172e4005bad34aafa51c32866/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-32B-FP8-dynamic model card, revision c6732fc26128341172e4005bad34aafa51c32866. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-32B-FP8-dynamic/blob/c6732fc26128341172e4005bad34aafa51c32866/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-32b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-32b-nvfp4",
    "name": "RedHatAI/Qwen3-32B-NVFP4 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-32B-NVFP4/blob/10a4cab9378ab938be865492b96ff42faff11c3f/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-32B-NVFP4 model card, revision 10a4cab9378ab938be865492b96ff42faff11c3f. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-32B-NVFP4/blob/10a4cab9378ab938be865492b96ff42faff11c3f/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-32b-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-32b-nvfp4a16",
    "name": "RedHatAI/Qwen3-32B-NVFP4A16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-32B-NVFP4A16/blob/c572e992f6042d2b6bfa6853f55b8196a4c8cf44/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-32B-NVFP4A16 model card, revision c572e992f6042d2b6bfa6853f55b8196a4c8cf44. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-32B-NVFP4A16/blob/c572e992f6042d2b6bfa6853f55b8196a4c8cf44/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-32b-nvfp4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-32b-w4a16",
    "name": "RedHatAI/Qwen3-32B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-32B-quantized.w4a16/blob/0364e6546d36447655a78bac8ab181572fd6ff24/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-32B-quantized.w4a16 model card, revision 0364e6546d36447655a78bac8ab181572fd6ff24. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-32B-quantized.w4a16/blob/0364e6546d36447655a78bac8ab181572fd6ff24/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-32b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-4b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-4B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-4B-FP8-dynamic/blob/31ad70c60365f34890b3118d38d1ad4dce9a6d99/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-4B-FP8-dynamic model card, revision 31ad70c60365f34890b3118d38d1ad4dce9a6d99. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-4B-FP8-dynamic/blob/31ad70c60365f34890b3118d38d1ad4dce9a6d99/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-4b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-4b-instruct-2507-w4a16",
    "name": "RedHatAI/Qwen3-4B-Instruct-2507-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-4B-Instruct-2507-quantized.w4a16/blob/5d4fde88f26ef2577a3e660db6304eaa6cbe99a1/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-4B-Instruct-2507-quantized.w4a16 model card, revision 5d4fde88f26ef2577a3e660db6304eaa6cbe99a1. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-4B-Instruct-2507-quantized.w4a16/blob/5d4fde88f26ef2577a3e660db6304eaa6cbe99a1/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-4b-instruct-2507-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-4b-instruct-2507-w8a8",
    "name": "RedHatAI/Qwen3-4B-Instruct-2507-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-4B-Instruct-2507-quantized.w8a8/blob/31ec5013b3319d7ba82b2a354d6e48b376204dd3/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-4B-Instruct-2507-quantized.w8a8 model card, revision 31ec5013b3319d7ba82b2a354d6e48b376204dd3. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-4B-Instruct-2507-quantized.w8a8/blob/31ec5013b3319d7ba82b2a354d6e48b376204dd3/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-4b-instruct-2507-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-4b-thinking-2507-w4a16",
    "name": "RedHatAI/Qwen3-4B-Thinking-2507-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-4B-Thinking-2507-quantized.w4a16/blob/6131664544d9d715c78f3b6c092fbe5568e2e79e/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-4B-Thinking-2507-quantized.w4a16 model card, revision 6131664544d9d715c78f3b6c092fbe5568e2e79e. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-4B-Thinking-2507-quantized.w4a16/blob/6131664544d9d715c78f3b6c092fbe5568e2e79e/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-4b-thinking-2507-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-4b-thinking-2507-w8a8",
    "name": "RedHatAI/Qwen3-4B-Thinking-2507-quantized.w8a8 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-4B-Thinking-2507-quantized.w8a8/blob/a6d3fbb1c3cd5c34ce1bdd4e244e04647f48691c/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-4B-Thinking-2507-quantized.w8a8 model card, revision a6d3fbb1c3cd5c34ce1bdd4e244e04647f48691c. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-4B-Thinking-2507-quantized.w8a8/blob/a6d3fbb1c3cd5c34ce1bdd4e244e04647f48691c/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-4b-thinking-2507-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-4b-w4a16",
    "name": "RedHatAI/Qwen3-4B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-4B-quantized.w4a16/blob/7a37568f5fcaf2ad3f84cdb9a1bb612546645614/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-4B-quantized.w4a16 model card, revision 7a37568f5fcaf2ad3f84cdb9a1bb612546645614. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-4B-quantized.w4a16/blob/7a37568f5fcaf2ad3f84cdb9a1bb612546645614/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-4b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-8b-fp8-block",
    "name": "RedHatAI/Qwen3-8B-FP8-block model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-8B-FP8-block/blob/b4fbea251a7905936ad2656faf78d4ca2839d9ad/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-8B-FP8-block model card, revision b4fbea251a7905936ad2656faf78d4ca2839d9ad. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-8B-FP8-block/blob/b4fbea251a7905936ad2656faf78d4ca2839d9ad/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-8b-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-8b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-8B-FP8-dynamic model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-8B-FP8-dynamic/blob/05233ce1e0565b5fdc9cfa000ab840152ed30c70/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-8B-FP8-dynamic model card, revision 05233ce1e0565b5fdc9cfa000ab840152ed30c70. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-8B-FP8-dynamic/blob/05233ce1e0565b5fdc9cfa000ab840152ed30c70/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 45: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-8b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-8b-nvfp4",
    "name": "RedHatAI/Qwen3-8B-NVFP4 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-8B-NVFP4/blob/e391349c110709b87bfc2ad2fde3f50dc5839fd8/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-8B-NVFP4 model card, revision e391349c110709b87bfc2ad2fde3f50dc5839fd8. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-8B-NVFP4/blob/e391349c110709b87bfc2ad2fde3f50dc5839fd8/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-8b-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-8b-w4a16",
    "name": "RedHatAI/Qwen3-8B-quantized.w4a16 model card (Red Hat AI quantized checkpoint evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-8B-quantized.w4a16/blob/32527053243a382bc56e941c964fc516a5014d39/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Qwen3-8B-quantized.w4a16 model card, revision 32527053243a382bc56e941c964fc516a5014d39. Selected evaluation cells (quantized and Red Hat-evaluated unquantized baseline) transcribed; percentages divided by 100 and retention ratios derived. Qwen models by Alibaba Cloud Qwen team. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-8B-quantized.w4a16/blob/32527053243a382bc56e941c964fc516a5014d39/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3: \"license: apache-2.0\" (no separate LICENSE/NOTICE file in the repository at this revision). Apache-2.0 covers the repository's documentation Source form; only Red Hat's own evaluation cells are used; linked vendor/third-party results are not reused.",
    "license_url": "/licenses/quant-redhatai-qwen3-8b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-235b-a22b-fp8-dynamic",
    "name": "RedHatAI/Qwen3-235B-A22B-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-235B-A22B-FP8-dynamic/blob/627b08fd72c21c3e0565d87764adce2d87427afc/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-235B-A22B-FP8-dynamic model card (Hugging Face, revision 627b08fd72c21c3e0565d87764adce2d87427afc), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-FP8-dynamic/blob/627b08fd72c21c3e0565d87764adce2d87427afc/README.md",
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-FP8-dynamic/blob/627b08fd72c21c3e0565d87764adce2d87427afc/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-235b-a22b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-235b-a22b-instruct-2507-nvfp4",
    "name": "RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4/blob/9514aa67dfafa8d459c0372d167597bc7b5811d3/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4 model card (Hugging Face, revision 9514aa67dfafa8d459c0372d167597bc7b5811d3), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4/blob/9514aa67dfafa8d459c0372d167597bc7b5811d3/README.md",
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4/blob/9514aa67dfafa8d459c0372d167597bc7b5811d3/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-235b-a22b-instruct-2507-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-235b-a22b-nvfp4",
    "name": "RedHatAI/Qwen3-235B-A22B-NVFP4 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-235B-A22B-NVFP4/blob/1b32bc2de6af4465acc7cbafa9d842a15d07457a/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-235B-A22B-NVFP4 model card (Hugging Face, revision 1b32bc2de6af4465acc7cbafa9d842a15d07457a), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-NVFP4/blob/1b32bc2de6af4465acc7cbafa9d842a15d07457a/README.md",
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-NVFP4/blob/1b32bc2de6af4465acc7cbafa9d842a15d07457a/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-235b-a22b-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-235b-a22b-instruct-2507-w8a8",
    "name": "RedHatAI/Qwen3-235B-A22B-Instruct-2507-quantized.w8a8 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-235B-A22B-Instruct-2507-quantized.w8a8/blob/3ac6f8cbd777cd3c5f855ae1d01fb416b050e66b/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-235B-A22B-Instruct-2507-quantized.w8a8 model card (Hugging Face, revision 3ac6f8cbd777cd3c5f855ae1d01fb416b050e66b), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-Instruct-2507-quantized.w8a8/blob/3ac6f8cbd777cd3c5f855ae1d01fb416b050e66b/README.md",
      "huggingface.co/RedHatAI/Qwen3-235B-A22B-Instruct-2507-quantized.w8a8/blob/3ac6f8cbd777cd3c5f855ae1d01fb416b050e66b/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-235b-a22b-instruct-2507-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-next-80b-a3b-instruct-fp8-block",
    "name": "RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-block model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-block/blob/ab944c4c176d211bfa625e8645d12282906761f9/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-block model card (Hugging Face, revision ab944c4c176d211bfa625e8645d12282906761f9), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-block/blob/ab944c4c176d211bfa625e8645d12282906761f9/README.md",
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-block/blob/ab944c4c176d211bfa625e8645d12282906761f9/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-next-80b-a3b-instruct-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-next-80b-a3b-instruct-fp8-dynamic",
    "name": "RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-dynamic/blob/fd5b264444ce2bb0686d4914965bae94e890f422/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-dynamic model card (Hugging Face, revision fd5b264444ce2bb0686d4914965bae94e890f422), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-dynamic/blob/fd5b264444ce2bb0686d4914965bae94e890f422/README.md",
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-FP8-dynamic/blob/fd5b264444ce2bb0686d4914965bae94e890f422/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-next-80b-a3b-instruct-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-next-80b-a3b-instruct-w4a16",
    "name": "RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w4a16 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w4a16/blob/ac9dc5b939ba948ab378b8638cfcce4ac4d5642b/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w4a16 model card (Hugging Face, revision ac9dc5b939ba948ab378b8638cfcce4ac4d5642b), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w4a16/blob/ac9dc5b939ba948ab378b8638cfcce4ac4d5642b/README.md",
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w4a16/blob/ac9dc5b939ba948ab378b8638cfcce4ac4d5642b/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-next-80b-a3b-instruct-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-next-80b-a3b-instruct-w8a8",
    "name": "RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8/blob/509f3bf6f13341cf4bbca584acd2ba020d09aaf2/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8 model card (Hugging Face, revision 509f3bf6f13341cf4bbca584acd2ba020d09aaf2), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8/blob/509f3bf6f13341cf4bbca584acd2ba020d09aaf2/README.md",
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Instruct-quantized.w8a8/blob/509f3bf6f13341cf4bbca584acd2ba020d09aaf2/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-next-80b-a3b-instruct-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-next-80b-a3b-thinking-fp8-dynamic",
    "name": "RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-dynamic/blob/c67f21402fe030480002fc69f23df9c7cba75bb5/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-dynamic model card (Hugging Face, revision c67f21402fe030480002fc69f23df9c7cba75bb5), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-dynamic/blob/c67f21402fe030480002fc69f23df9c7cba75bb5/README.md",
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-dynamic/blob/c67f21402fe030480002fc69f23df9c7cba75bb5/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-next-80b-a3b-thinking-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-next-80b-a3b-thinking-w4a16",
    "name": "RedHatAI/Qwen3-Next-80B-A3B-Thinking-quantized.w4a16 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-quantized.w4a16/blob/76f7498176afa06571650e00dd3f423f608c4369/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-Next-80B-A3B-Thinking-quantized.w4a16 model card (Hugging Face, revision 76f7498176afa06571650e00dd3f423f608c4369), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-quantized.w4a16/blob/76f7498176afa06571650e00dd3f423f608c4369/README.md",
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-quantized.w4a16/blob/76f7498176afa06571650e00dd3f423f608c4369/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-next-80b-a3b-thinking-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-next-80b-a3b-thinking-fp8-block",
    "name": "RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-block model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-block/blob/30b4ae8ac43a84fa8b328ad8b1c64adb7070867f/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-block model card (Hugging Face, revision 30b4ae8ac43a84fa8b328ad8b1c64adb7070867f), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-block/blob/30b4ae8ac43a84fa8b328ad8b1c64adb7070867f/README.md",
      "huggingface.co/RedHatAI/Qwen3-Next-80B-A3B-Thinking-FP8-block/blob/30b4ae8ac43a84fa8b328ad8b1c64adb7070867f/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-next-80b-a3b-thinking-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-122b-a10b-fp8-dynamic",
    "name": "RedHatAI/Qwen3.5-122B-A10B-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-122B-A10B-FP8-dynamic/blob/dae17507e035bac7e87c3daaf3939656a6550784/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-122B-A10B-FP8-dynamic model card (Hugging Face, revision dae17507e035bac7e87c3daaf3939656a6550784), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-122B-A10B-FP8-dynamic/blob/dae17507e035bac7e87c3daaf3939656a6550784/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-122B-A10B-FP8-dynamic/blob/dae17507e035bac7e87c3daaf3939656a6550784/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-122b-a10b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-35b-a3b-fp8-dynamic",
    "name": "RedHatAI/Qwen3.5-35B-A3B-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-35B-A3B-FP8-dynamic/blob/5448394de7a7f81eb5e92f9608aa91a14bd11ceb/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-35B-A3B-FP8-dynamic model card (Hugging Face, revision 5448394de7a7f81eb5e92f9608aa91a14bd11ceb), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-35B-A3B-FP8-dynamic/blob/5448394de7a7f81eb5e92f9608aa91a14bd11ceb/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-35B-A3B-FP8-dynamic/blob/5448394de7a7f81eb5e92f9608aa91a14bd11ceb/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-35b-a3b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-397b-a17b-fp8-dynamic",
    "name": "RedHatAI/Qwen3.5-397B-A17B-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-397B-A17B-FP8-dynamic/blob/b75204601a4140423d5107fdccd6b928cfb1d7c5/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-397B-A17B-FP8-dynamic model card (Hugging Face, revision b75204601a4140423d5107fdccd6b928cfb1d7c5), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-397B-A17B-FP8-dynamic/blob/b75204601a4140423d5107fdccd6b928cfb1d7c5/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-397B-A17B-FP8-dynamic/blob/b75204601a4140423d5107fdccd6b928cfb1d7c5/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-397b-a17b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-4b-fp8-dynamic",
    "name": "RedHatAI/Qwen3.5-4B-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-4B-FP8-dynamic/blob/397b7ba47a99b3221ebfc0cfc5a279118cb733ad/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-4B-FP8-dynamic model card (Hugging Face, revision 397b7ba47a99b3221ebfc0cfc5a279118cb733ad), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-4B-FP8-dynamic/blob/397b7ba47a99b3221ebfc0cfc5a279118cb733ad/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-4B-FP8-dynamic/blob/397b7ba47a99b3221ebfc0cfc5a279118cb733ad/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-4b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-4b-w4a16",
    "name": "RedHatAI/Qwen3.5-4B-quantized.w4a16 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-4B-quantized.w4a16/blob/7a613872f394578b0b52b683ff4ac47516b4bcaf/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-4B-quantized.w4a16 model card (Hugging Face, revision 7a613872f394578b0b52b683ff4ac47516b4bcaf), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-4B-quantized.w4a16/blob/7a613872f394578b0b52b683ff4ac47516b4bcaf/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-4B-quantized.w4a16/blob/7a613872f394578b0b52b683ff4ac47516b4bcaf/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-4b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-4b-w8a8",
    "name": "RedHatAI/Qwen3.5-4B-quantized.w8a8 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-4B-quantized.w8a8/blob/c30ff438442b337db924cadb5136847bbf3fb95a/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-4B-quantized.w8a8 model card (Hugging Face, revision c30ff438442b337db924cadb5136847bbf3fb95a), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-4B-quantized.w8a8/blob/c30ff438442b337db924cadb5136847bbf3fb95a/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-4B-quantized.w8a8/blob/c30ff438442b337db924cadb5136847bbf3fb95a/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-4b-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-9b-fp8-dynamic",
    "name": "RedHatAI/Qwen3.5-9B-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-9B-FP8-dynamic/blob/790f0576d2d77dd5322aa0603a470bd9e3a3d1f6/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-9B-FP8-dynamic model card (Hugging Face, revision 790f0576d2d77dd5322aa0603a470bd9e3a3d1f6), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-9B-FP8-dynamic/blob/790f0576d2d77dd5322aa0603a470bd9e3a3d1f6/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-9B-FP8-dynamic/blob/790f0576d2d77dd5322aa0603a470bd9e3a3d1f6/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-9b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-9b-w4a16",
    "name": "RedHatAI/Qwen3.5-9B-quantized.w4a16 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-9B-quantized.w4a16/blob/a398088c4228b0ae0c8c78df88fd1e4bf445f068/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-9B-quantized.w4a16 model card (Hugging Face, revision a398088c4228b0ae0c8c78df88fd1e4bf445f068), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-9B-quantized.w4a16/blob/a398088c4228b0ae0c8c78df88fd1e4bf445f068/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-9B-quantized.w4a16/blob/a398088c4228b0ae0c8c78df88fd1e4bf445f068/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-9b-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-5-9b-w8a8",
    "name": "RedHatAI/Qwen3.5-9B-quantized.w8a8 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.5-9B-quantized.w8a8/blob/4257ccf0e5c6f87246fd96ea2e4fb16817101558/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.5-9B-quantized.w8a8 model card (Hugging Face, revision 4257ccf0e5c6f87246fd96ea2e4fb16817101558), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.5-9B-quantized.w8a8/blob/4257ccf0e5c6f87246fd96ea2e4fb16817101558/README.md",
      "huggingface.co/RedHatAI/Qwen3.5-9B-quantized.w8a8/blob/4257ccf0e5c6f87246fd96ea2e4fb16817101558/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-5-9b-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-6-35b-a3b-fp8",
    "name": "RedHatAI/Qwen3.6-35B-A3B-FP8 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-FP8/blob/ea247061a1c961ddc12daff63a326b345a0489b4/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.6-35B-A3B-FP8 model card (Hugging Face, revision ea247061a1c961ddc12daff63a326b345a0489b4), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.6-35B-A3B-FP8/blob/ea247061a1c961ddc12daff63a326b345a0489b4/README.md",
      "huggingface.co/RedHatAI/Qwen3.6-35B-A3B-FP8/blob/ea247061a1c961ddc12daff63a326b345a0489b4/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\" and the repository contains an Apache License 2.0 LICENSE file at the pinned revision. Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-6-35b-a3b-fp8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-6-35b-a3b-fp8-dynamic",
    "name": "RedHatAI/Qwen3.6-35B-A3B-FP8-dynamic model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-FP8-dynamic/blob/6496d54f9127add4e6a62b7c5e3c5579ddb04ac7/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.6-35B-A3B-FP8-dynamic model card (Hugging Face, revision 6496d54f9127add4e6a62b7c5e3c5579ddb04ac7), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.6-35B-A3B-FP8-dynamic/blob/6496d54f9127add4e6a62b7c5e3c5579ddb04ac7/README.md",
      "huggingface.co/RedHatAI/Qwen3.6-35B-A3B-FP8-dynamic/blob/6496d54f9127add4e6a62b7c5e3c5579ddb04ac7/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-6-35b-a3b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-6-35b-a3b-nvfp4",
    "name": "RedHatAI/Qwen3.6-35B-A3B-NVFP4 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4/blob/e94a930c8e9b8a7d3763e449289c261c9c7d3e5e/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.6-35B-A3B-NVFP4 model card (Hugging Face, revision e94a930c8e9b8a7d3763e449289c261c9c7d3e5e), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4/blob/e94a930c8e9b8a7d3763e449289c261c9c7d3e5e/README.md",
      "huggingface.co/RedHatAI/Qwen3.6-35B-A3B-NVFP4/blob/e94a930c8e9b8a7d3763e449289c261c9c7d3e5e/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-6-35b-a3b-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-8-27b-int4",
    "name": "RedHatAI/Qwen3.8-27B-INT4 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.8-27B-INT4/blob/2f0fa48fea4949b56685ddfc48c05dd81c7d772b/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.8-27B-INT4 model card (Hugging Face, revision 2f0fa48fea4949b56685ddfc48c05dd81c7d772b), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.8-27B-INT4/blob/2f0fa48fea4949b56685ddfc48c05dd81c7d772b/README.md",
      "huggingface.co/RedHatAI/Qwen3.8-27B-INT4/blob/2f0fa48fea4949b56685ddfc48c05dd81c7d772b/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-8-27b-int4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-8-27b-mxfp4",
    "name": "RedHatAI/Qwen3.8-27B-MXFP4 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.8-27B-MXFP4/blob/651eed99386dae5645cd4949024e8ebd163a533b/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.8-27B-MXFP4 model card (Hugging Face, revision 651eed99386dae5645cd4949024e8ebd163a533b), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.8-27B-MXFP4/blob/651eed99386dae5645cd4949024e8ebd163a533b/README.md",
      "huggingface.co/RedHatAI/Qwen3.8-27B-MXFP4/blob/651eed99386dae5645cd4949024e8ebd163a533b/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-8-27b-mxfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwen3-8-27b-nvfp4",
    "name": "RedHatAI/Qwen3.8-27B-NVFP4 model card (Red Hat quantized-model evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Qwen3.8-27B-NVFP4/blob/d23c6ff198a005532746610e9d719c9f6d27a2b1/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat, Inc. / Neural Magic, RedHatAI/Qwen3.8-27B-NVFP4 model card (Hugging Face, revision d23c6ff198a005532746610e9d719c9f6d27a2b1), Apache-2.0. Selected accuracy cells transcribed; percentages divided by 100; retention ratios recomputed; recipe labels normalized. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Qwen3.8-27B-NVFP4/blob/d23c6ff198a005532746610e9d719c9f6d27a2b1/README.md",
      "huggingface.co/RedHatAI/Qwen3.8-27B-NVFP4/blob/d23c6ff198a005532746610e9d719c9f6d27a2b1/config.json"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata declares \"license: apache-2.0\"; no LICENSE file in the repository, so the explicit card declaration is relied on (same basis as the accepted card-final-ling-3-0-flash precedent). Apache-2.0 s.1 \"Source\" form includes documentation; s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce, prepare Derivative Works of, publicly display and distribute; s.4 requires retaining the licence and marking changes. Only Red Hat's own accuracy measurements are used.",
    "license_url": "/licenses/quant-redhatai-qwen3-8-27b-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-1-5b-fp8-dynamic",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-FP8-dynamic/blob/b87f998fcda08540b81b50844bcdb811540db5e8/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-FP8-dynamic model card, revision b87f998fcda08540b81b50844bcdb811540db5e8. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-FP8-dynamic/blob/b87f998fcda08540b81b50844bcdb811540db5e8/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-1-5b-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-1-5b-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-quantized.w4a16/blob/ececd8aa3863503391260ef30bec33887694dea0/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-quantized.w4a16 model card, revision ececd8aa3863503391260ef30bec33887694dea0. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-1.5B-quantized.w4a16/blob/ececd8aa3863503391260ef30bec33887694dea0/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-1-5b-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-7b-fp8-dynamic",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-7B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-7B-FP8-dynamic/blob/ceb2fcd178d477cd21b92ffb43164c74165a212c/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-7B-FP8-dynamic model card, revision ceb2fcd178d477cd21b92ffb43164c74165a212c. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-7B-FP8-dynamic/blob/ceb2fcd178d477cd21b92ffb43164c74165a212c/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-7b-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-7b-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w4a16/blob/83b68401b8a9f9439725acc32a6e9c350451fb64/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w4a16 model card, revision 83b68401b8a9f9439725acc32a6e9c350451fb64. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w4a16/blob/83b68401b8a9f9439725acc32a6e9c350451fb64/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-7b-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-7b-quantized-w8a8",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w8a8/blob/925259de674c80799e9455bfcd2bf50f999022cb/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w8a8 model card, revision 925259de674c80799e9455bfcd2bf50f999022cb. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-7B-quantized.w8a8/blob/925259de674c80799e9455bfcd2bf50f999022cb/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-7b-quantized-w8a8-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-14b-fp8-dynamic",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-14B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-14B-FP8-dynamic/blob/cdf473db71c978fcf36cbdacd44da74ebd055354/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-14B-FP8-dynamic model card, revision cdf473db71c978fcf36cbdacd44da74ebd055354. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-14B-FP8-dynamic/blob/cdf473db71c978fcf36cbdacd44da74ebd055354/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-14b-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-14b-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w4a16/blob/90ab462f58256472697848376961c8a5b754b094/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w4a16 model card, revision 90ab462f58256472697848376961c8a5b754b094. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w4a16/blob/90ab462f58256472697848376961c8a5b754b094/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-14b-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-14b-quantized-w8a8",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w8a8/blob/52aeeeebff0c6a91a04a2971df4e106dbaf94966/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w8a8 model card, revision 52aeeeebff0c6a91a04a2971df4e106dbaf94966. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-14B-quantized.w8a8/blob/52aeeeebff0c6a91a04a2971df4e106dbaf94966/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-14b-quantized-w8a8-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-32b-fp8-dynamic",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-32B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-32B-FP8-dynamic/blob/e702fff1b22be372c8893745aa9a13cf1a5d0d8e/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-32B-FP8-dynamic model card, revision e702fff1b22be372c8893745aa9a13cf1a5d0d8e. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-32B-FP8-dynamic/blob/e702fff1b22be372c8893745aa9a13cf1a5d0d8e/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-32b-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-32b-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w4a16/blob/f3572d23e9bc8dc194f361cfaf9ab73383e9125a/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w4a16 model card, revision f3572d23e9bc8dc194f361cfaf9ab73383e9125a. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w4a16/blob/f3572d23e9bc8dc194f361cfaf9ab73383e9125a/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-32b-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-qwen-32b-quantized-w8a8",
    "name": "RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8/blob/bfc01211470299398548c5d6ff8ca2e48dc9a3dc/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8 model card, revision bfc01211470299398548c5d6ff8ca2e48dc9a3dc. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Qwen-32B-quantized.w8a8/blob/bfc01211470299398548c5d6ff8ca2e48dc9a3dc/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-qwen-32b-quantized-w8a8-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-llama-8b-fp8-dynamic",
    "name": "RedHatAI/DeepSeek-R1-Distill-Llama-8B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-8B-FP8-dynamic/blob/5d548d9169c53b7bb7ef0b3bc261509d5da6e3dd/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Llama-8B-FP8-dynamic model card, revision 5d548d9169c53b7bb7ef0b3bc261509d5da6e3dd. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-8B-FP8-dynamic/blob/5d548d9169c53b7bb7ef0b3bc261509d5da6e3dd/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. Weights derive from Llama 3.x and remain subject to their upstream licence; the MIT declaration is relied on only for Red Hat's own card text and evaluation numbers.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-llama-8b-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-llama-8b-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w4a16/blob/2ed7bf0c3a423fd114e70a86f7e30420870709af/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w4a16 model card, revision 2ed7bf0c3a423fd114e70a86f7e30420870709af. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w4a16/blob/2ed7bf0c3a423fd114e70a86f7e30420870709af/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. Weights derive from Llama 3.x and remain subject to their upstream licence; the MIT declaration is relied on only for Red Hat's own card text and evaluation numbers.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-llama-8b-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-llama-8b-quantized-w8a8",
    "name": "RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w8a8/blob/d167aa19e86a90b8f205cebd68e986a72c76be57/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w8a8 model card, revision d167aa19e86a90b8f205cebd68e986a72c76be57. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-8B-quantized.w8a8/blob/d167aa19e86a90b8f205cebd68e986a72c76be57/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. Weights derive from Llama 3.x and remain subject to their upstream licence; the MIT declaration is relied on only for Red Hat's own card text and evaluation numbers.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-llama-8b-quantized-w8a8-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-llama-70b-fp8-dynamic",
    "name": "RedHatAI/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic/blob/fb3637a1165cec3832958bd72ebbe04021601489/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic model card, revision fb3637a1165cec3832958bd72ebbe04021601489. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-70B-FP8-dynamic/blob/fb3637a1165cec3832958bd72ebbe04021601489/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. Weights derive from Llama 3.x and remain subject to their upstream licence; the MIT declaration is relied on only for Red Hat's own card text and evaluation numbers.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-llama-70b-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-llama-70b-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w4a16/blob/b5b2803148745cd6986a134393fea571a28af1d5/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w4a16 model card, revision b5b2803148745cd6986a134393fea571a28af1d5. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w4a16/blob/b5b2803148745cd6986a134393fea571a28af1d5/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. Weights derive from Llama 3.x and remain subject to their upstream licence; the MIT declaration is relied on only for Red Hat's own card text and evaluation numbers.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-llama-70b-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-distill-llama-70b-quantized-w8a8",
    "name": "RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w8a8/blob/2fb57dc0dc61ccf5b282f96e9a6b939b447ecdbf/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w8a8 model card, revision 2fb57dc0dc61ccf5b282f96e9a6b939b447ecdbf. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w8a8/blob/2fb57dc0dc61ccf5b282f96e9a6b939b447ecdbf/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. Weights derive from Llama 3.x and remain subject to their upstream licence; the MIT declaration is relied on only for Red Hat's own card text and evaluation numbers.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-distill-llama-70b-quantized-w8a8-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-quantized.w4a16/blob/12a44c05b1b94f9759420d9acc05328921f76dde/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-quantized.w4a16 model card, revision 12a44c05b1b94f9759420d9acc05328921f76dde. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-quantized.w4a16/blob/12a44c05b1b94f9759420d9acc05328921f76dde/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-deepseek-r1-0528-quantized-w4a16",
    "name": "RedHatAI/DeepSeek-R1-0528-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/DeepSeek-R1-0528-quantized.w4a16/blob/0efe34e82e4612e726c42f6fd44116e0244b33f8/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/DeepSeek-R1-0528-quantized.w4a16 model card, revision 0efe34e82e4612e726c42f6fd44116e0244b33f8. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/DeepSeek-R1-0528-quantized.w4a16/blob/0efe34e82e4612e726c42f6fd44116e0244b33f8/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 18 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-deepseek-r1-0528-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-devstral-small-2507-fp8-dynamic",
    "name": "RedHatAI/Devstral-Small-2507-FP8-Dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/Devstral-Small-2507-FP8-Dynamic/blob/0566e03128c6190f34931e750653c2f5334174d7/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Devstral-Small-2507-FP8-Dynamic model card, revision 0566e03128c6190f34931e750653c2f5334174d7. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Devstral-Small-2507-FP8-Dynamic/blob/0566e03128c6190f34931e750653c2f5334174d7/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-devstral-small-2507-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-devstral-small-2507-quantized-w4a16",
    "name": "RedHatAI/Devstral-Small-2507-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/Devstral-Small-2507-quantized.w4a16/blob/bb96f0d1a876d6f6f3ed657b6a965328697ba2f9/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Devstral-Small-2507-quantized.w4a16 model card, revision bb96f0d1a876d6f6f3ed657b6a965328697ba2f9. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Devstral-Small-2507-quantized.w4a16/blob/bb96f0d1a876d6f6f3ed657b6a965328697ba2f9/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-devstral-small-2507-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-devstral-small-2507-quantized-w8a8",
    "name": "RedHatAI/Devstral-Small-2507-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/Devstral-Small-2507-quantized.w8a8/blob/b152b0574658cdd159721472ebf5353d8d199451/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Devstral-Small-2507-quantized.w8a8 model card, revision b152b0574658cdd159721472ebf5353d8d199451. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Devstral-Small-2507-quantized.w8a8/blob/b152b0574658cdd159721472ebf5353d8d199451/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-devstral-small-2507-quantized-w8a8-MIT.txt"
  },
  {
    "id": "quant-redhatai-ministral-3-14b-instruct-2512-fp8-dynamic",
    "name": "RedHatAI/Ministral-3-14B-Instruct-2512-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Ministral-3-14B-Instruct-2512-FP8-dynamic/blob/c5d7753e839c659e04906bcfce7eb6ea50d6d18c/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Ministral-3-14B-Instruct-2512-FP8-dynamic model card, revision c5d7753e839c659e04906bcfce7eb6ea50d6d18c. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Ministral-3-14B-Instruct-2512-FP8-dynamic/blob/c5d7753e839c659e04906bcfce7eb6ea50d6d18c/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-ministral-3-14b-instruct-2512-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-ministral-3-14b-instruct-2512-nvfp4",
    "name": "RedHatAI/Ministral-3-14B-Instruct-2512-NVFP4 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Ministral-3-14B-Instruct-2512-NVFP4/blob/ae31e2b3d5aca78b8721e071291d7feed5a87dad/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Ministral-3-14B-Instruct-2512-NVFP4 model card, revision ae31e2b3d5aca78b8721e071291d7feed5a87dad. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Ministral-3-14B-Instruct-2512-NVFP4/blob/ae31e2b3d5aca78b8721e071291d7feed5a87dad/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-ministral-3-14b-instruct-2512-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-mistral-small-3-2-24b-instruct-2506-nvfp4",
    "name": "RedHatAI/Mistral-Small-3.2-24B-Instruct-2506-NVFP4 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Mistral-Small-3.2-24B-Instruct-2506-NVFP4/blob/5102662a785ad4f2c47eb54624bf3132ee620a58/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Mistral-Small-3.2-24B-Instruct-2506-NVFP4 model card, revision 5102662a785ad4f2c47eb54624bf3132ee620a58. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Mistral-Small-3.2-24B-Instruct-2506-NVFP4/blob/5102662a785ad4f2c47eb54624bf3132ee620a58/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-mistral-small-3-2-24b-instruct-2506-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-phi-4-fp8-dynamic",
    "name": "RedHatAI/phi-4-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/phi-4-FP8-dynamic/blob/8f62f4b85ccc1c7fda9f178976c1162f37e43651/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/phi-4-FP8-dynamic model card, revision 8f62f4b85ccc1c7fda9f178976c1162f37e43651. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/phi-4-FP8-dynamic/blob/8f62f4b85ccc1c7fda9f178976c1162f37e43651/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 21 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-phi-4-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-phi-4-quantized-w4a16",
    "name": "RedHatAI/phi-4-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/phi-4-quantized.w4a16/blob/e05cd2e17a849b824c85d6acdec3f62be9cd24c1/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/phi-4-quantized.w4a16 model card, revision e05cd2e17a849b824c85d6acdec3f62be9cd24c1. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/phi-4-quantized.w4a16/blob/e05cd2e17a849b824c85d6acdec3f62be9cd24c1/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 22 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-phi-4-quantized-w4a16-MIT.txt"
  },
  {
    "id": "quant-redhatai-phi-4-quantized-w8a8",
    "name": "RedHatAI/phi-4-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/phi-4-quantized.w8a8/blob/47ae81515aec82176f334e02506eea10dacfddca/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/phi-4-quantized.w8a8 model card, revision 47ae81515aec82176f334e02506eea10dacfddca. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/phi-4-quantized.w8a8/blob/47ae81515aec82176f334e02506eea10dacfddca/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 22 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-phi-4-quantized-w8a8-MIT.txt"
  },
  {
    "id": "quant-redhatai-phi-4-mini-instruct-fp8-dynamic",
    "name": "RedHatAI/Phi-4-mini-instruct-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/Phi-4-mini-instruct-FP8-dynamic/blob/5450d52149bf44e81cbbd30b2f02762947845bb3/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Phi-4-mini-instruct-FP8-dynamic model card, revision 5450d52149bf44e81cbbd30b2f02762947845bb3. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Phi-4-mini-instruct-FP8-dynamic/blob/5450d52149bf44e81cbbd30b2f02762947845bb3/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 27 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-phi-4-mini-instruct-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-phi-4-reasoning-fp8-dynamic",
    "name": "RedHatAI/Phi-4-reasoning-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/Phi-4-reasoning-FP8-dynamic/blob/10256f614f546d20fddd470592e55e59e6a6e2e4/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Phi-4-reasoning-FP8-dynamic model card, revision 10256f614f546d20fddd470592e55e59e6a6e2e4. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Phi-4-reasoning-FP8-dynamic/blob/10256f614f546d20fddd470592e55e59e6a6e2e4/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-phi-4-reasoning-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-qwq-32b-fp8-dynamic",
    "name": "RedHatAI/QwQ-32B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://huggingface.co/RedHatAI/QwQ-32B-FP8-dynamic/blob/c4f1385172c9b31fd151820312aedd6a20fe6830/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/QwQ-32B-FP8-dynamic model card, revision c4f1385172c9b31fd151820312aedd6a20fe6830. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/QwQ-32B-FP8-dynamic/blob/c4f1385172c9b31fd151820312aedd6a20fe6830/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: mit`; no separate LICENSE file in the repository at this revision. The standard MIT grant expressly covers associated documentation files and permits commercial copying, publication and distribution subject to retaining the copyright and permission notice. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. The upstream Qwen/QwQ-32B repository is Apache-2.0; this Red Hat repository itself declares MIT and that declaration is relied on for the card.",
    "license_url": "/licenses/quant-redhatai-qwq-32b-fp8-dynamic-MIT.txt"
  },
  {
    "id": "quant-redhatai-qwq-32b-preview-fp8-dynamic",
    "name": "RedHatAI/QwQ-32B-Preview-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/QwQ-32B-Preview-FP8-dynamic/blob/de33103179314a2c626b998043238ce21700e64d/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/QwQ-32B-Preview-FP8-dynamic model card, revision de33103179314a2c626b998043238ce21700e64d. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/QwQ-32B-Preview-FP8-dynamic/blob/de33103179314a2c626b998043238ce21700e64d/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 5 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-qwq-32b-preview-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwq-32b-preview-quantized-w4a16",
    "name": "RedHatAI/QwQ-32B-Preview-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/QwQ-32B-Preview-quantized.w4a16/blob/be9a600807f053d20438b4511466d8eb0fef909c/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/QwQ-32B-Preview-quantized.w4a16 model card, revision be9a600807f053d20438b4511466d8eb0fef909c. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/QwQ-32B-Preview-quantized.w4a16/blob/be9a600807f053d20438b4511466d8eb0fef909c/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 6 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-qwq-32b-preview-quantized-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-qwq-32b-preview-quantized-w8a8",
    "name": "RedHatAI/QwQ-32B-Preview-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/QwQ-32B-Preview-quantized.w8a8/blob/bf17e01cb466e31ac49d5ae188e0b77121945cf9/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/QwQ-32B-Preview-quantized.w8a8 model card, revision bf17e01cb466e31ac49d5ae188e0b77121945cf9. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/QwQ-32B-Preview-quantized.w8a8/blob/bf17e01cb466e31ac49d5ae188e0b77121945cf9/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 6 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-qwq-32b-preview-quantized-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-smollm3-3b-fp8-dynamic",
    "name": "RedHatAI/SmolLM3-3B-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/SmolLM3-3B-FP8-dynamic/blob/01a88bc017932dbe4c7f0a4fc1db0bb72a7ab084/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/SmolLM3-3B-FP8-dynamic model card, revision 01a88bc017932dbe4c7f0a4fc1db0bb72a7ab084. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/SmolLM3-3B-FP8-dynamic/blob/01a88bc017932dbe4c7f0a4fc1db0bb72a7ab084/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 20 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-smollm3-3b-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-smollm3-3b-quantized-w4a16",
    "name": "RedHatAI/SmolLM3-3B-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/SmolLM3-3B-quantized.w4a16/blob/8d782c9fd7495f5d2add92b395d218b906e67678/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/SmolLM3-3B-quantized.w4a16 model card, revision 8d782c9fd7495f5d2add92b395d218b906e67678. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/SmolLM3-3B-quantized.w4a16/blob/8d782c9fd7495f5d2add92b395d218b906e67678/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 3 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-smollm3-3b-quantized-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-gemma-4-26b-a4b-it-fp8-dynamic",
    "name": "RedHatAI/gemma-4-26B-A4B-it-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/gemma-4-26B-A4B-it-FP8-dynamic/blob/ed35d7abe5d940da41b4ff06eb482feb0be8cb44/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/gemma-4-26B-A4B-it-FP8-dynamic model card, revision ed35d7abe5d940da41b4ff06eb482feb0be8cb44. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/gemma-4-26B-A4B-it-FP8-dynamic/blob/ed35d7abe5d940da41b4ff06eb482feb0be8cb44/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 8 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. The metadata license_link points to https://ai.google.dev/gemma/docs/gemma_4_license, which (checked 2026-09-25) is the verbatim Apache License 2.0; Gemma 4 is not under the earlier Gemma Terms of Use.",
    "license_url": "/licenses/quant-redhatai-gemma-4-26b-a4b-it-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-gemma-4-26b-a4b-it-nvfp4",
    "name": "RedHatAI/gemma-4-26B-A4B-it-NVFP4 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/gemma-4-26B-A4B-it-NVFP4/blob/5557756b8dce33ac72f2bd702b11729fdba3b839/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/gemma-4-26B-A4B-it-NVFP4 model card, revision 5557756b8dce33ac72f2bd702b11729fdba3b839. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/gemma-4-26B-A4B-it-NVFP4/blob/5557756b8dce33ac72f2bd702b11729fdba3b839/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 9 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. The metadata license_link points to https://ai.google.dev/gemma/docs/gemma_4_license, which (checked 2026-09-25) is the verbatim Apache License 2.0; Gemma 4 is not under the earlier Gemma Terms of Use.",
    "license_url": "/licenses/quant-redhatai-gemma-4-26b-a4b-it-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-gemma-4-31b-it-fp8-block",
    "name": "RedHatAI/gemma-4-31B-it-FP8-block model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/gemma-4-31B-it-FP8-block/blob/d7242548c457ab4b45bd3adb8937f2659af4739d/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/gemma-4-31B-it-FP8-block model card, revision d7242548c457ab4b45bd3adb8937f2659af4739d. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/gemma-4-31B-it-FP8-block/blob/d7242548c457ab4b45bd3adb8937f2659af4739d/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 8 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. The metadata license_link points to https://ai.google.dev/gemma/docs/gemma_4_license, which (checked 2026-09-25) is the verbatim Apache License 2.0; Gemma 4 is not under the earlier Gemma Terms of Use.",
    "license_url": "/licenses/quant-redhatai-gemma-4-31b-it-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-gemma-4-31b-it-fp8-dynamic",
    "name": "RedHatAI/gemma-4-31B-it-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/gemma-4-31B-it-FP8-dynamic/blob/d4ab4f579dd3516f97d8a6a4c98d0653480bad15/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/gemma-4-31B-it-FP8-dynamic model card, revision d4ab4f579dd3516f97d8a6a4c98d0653480bad15. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/gemma-4-31B-it-FP8-dynamic/blob/d4ab4f579dd3516f97d8a6a4c98d0653480bad15/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 8 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. The metadata license_link points to https://ai.google.dev/gemma/docs/gemma_4_license, which (checked 2026-09-25) is the verbatim Apache License 2.0; Gemma 4 is not under the earlier Gemma Terms of Use.",
    "license_url": "/licenses/quant-redhatai-gemma-4-31b-it-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-gemma-4-31b-it-nvfp4",
    "name": "RedHatAI/gemma-4-31B-it-NVFP4 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/gemma-4-31B-it-NVFP4/blob/5df75ccad9b74546f5be758a2a68acc2e16bbc9b/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/gemma-4-31B-it-NVFP4 model card, revision 5df75ccad9b74546f5be758a2a68acc2e16bbc9b. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/gemma-4-31B-it-NVFP4/blob/5df75ccad9b74546f5be758a2a68acc2e16bbc9b/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 8 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not. The metadata license_link points to https://ai.google.dev/gemma/docs/gemma_4_license, which (checked 2026-09-25) is the verbatim Apache License 2.0; Gemma 4 is not under the earlier Gemma Terms of Use.",
    "license_url": "/licenses/quant-redhatai-gemma-4-31b-it-nvfp4-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-granite-4-0-h-small-fp8-block",
    "name": "RedHatAI/granite-4.0-h-small-FP8-block model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/granite-4.0-h-small-FP8-block/blob/d65f13db6dee2495b3c8e32339cca3e616534e23/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/granite-4.0-h-small-FP8-block model card, revision d65f13db6dee2495b3c8e32339cca3e616534e23. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/granite-4.0-h-small-FP8-block/blob/d65f13db6dee2495b3c8e32339cca3e616534e23/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-granite-4-0-h-small-fp8-block-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-granite-4-0-h-small-fp8-dynamic",
    "name": "RedHatAI/granite-4.0-h-small-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/granite-4.0-h-small-FP8-dynamic/blob/440b960193f3e871ebf74f942e630fbbd5012602/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/granite-4.0-h-small-FP8-dynamic model card, revision 440b960193f3e871ebf74f942e630fbbd5012602. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/granite-4.0-h-small-FP8-dynamic/blob/440b960193f3e871ebf74f942e630fbbd5012602/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-granite-4-0-h-small-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-granite-4-0-h-tiny-fp8-dynamic",
    "name": "RedHatAI/granite-4.0-h-tiny-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/granite-4.0-h-tiny-FP8-dynamic/blob/85b19401d507a7d3ddfd93065dd5adf68ecc46a1/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/granite-4.0-h-tiny-FP8-dynamic model card, revision 85b19401d507a7d3ddfd93065dd5adf68ecc46a1. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/granite-4.0-h-tiny-FP8-dynamic/blob/85b19401d507a7d3ddfd93065dd5adf68ecc46a1/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-granite-4-0-h-tiny-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-apertus-70b-instruct-2509-fp8-dynamic",
    "name": "RedHatAI/Apertus-70B-Instruct-2509-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Apertus-70B-Instruct-2509-FP8-dynamic/blob/79a917d6a8034910fa610b097dc61e96eac65d29/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Apertus-70B-Instruct-2509-FP8-dynamic model card, revision 79a917d6a8034910fa610b097dc61e96eac65d29. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Apertus-70B-Instruct-2509-FP8-dynamic/blob/79a917d6a8034910fa610b097dc61e96eac65d29/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-apertus-70b-instruct-2509-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-apertus-8b-instruct-2509-fp8-dynamic",
    "name": "RedHatAI/Apertus-8B-Instruct-2509-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Apertus-8B-Instruct-2509-FP8-dynamic/blob/e9b17a517caa151279eeaf05fc973b3a11d9e0f9/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Apertus-8B-Instruct-2509-FP8-dynamic model card, revision e9b17a517caa151279eeaf05fc973b3a11d9e0f9. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Apertus-8B-Instruct-2509-FP8-dynamic/blob/e9b17a517caa151279eeaf05fc973b3a11d9e0f9/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 2 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-apertus-8b-instruct-2509-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-whisper-large-v3-fp8-dynamic",
    "name": "RedHatAI/whisper-large-v3-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/whisper-large-v3-FP8-dynamic/blob/a972f9c7c573f57054608ca4c47594f68da64031/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/whisper-large-v3-FP8-dynamic model card, revision a972f9c7c573f57054608ca4c47594f68da64031. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/whisper-large-v3-FP8-dynamic/blob/a972f9c7c573f57054608ca4c47594f68da64031/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 6 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-whisper-large-v3-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-whisper-large-v3-quantized-w8a8",
    "name": "RedHatAI/whisper-large-v3-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/whisper-large-v3-quantized.w8a8/blob/11d614fb6fc82faeaa7414b0e5613e2fc37768a5/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/whisper-large-v3-quantized.w8a8 model card, revision 11d614fb6fc82faeaa7414b0e5613e2fc37768a5. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/whisper-large-v3-quantized.w8a8/blob/11d614fb6fc82faeaa7414b0e5613e2fc37768a5/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 7 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-whisper-large-v3-quantized-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-whisper-large-v3-turbo-fp8-dynamic",
    "name": "RedHatAI/whisper-large-v3-turbo-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/whisper-large-v3-turbo-FP8-dynamic/blob/e72a6dca29d039a5c9ea13e622e496ca61e85c34/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/whisper-large-v3-turbo-FP8-dynamic model card, revision e72a6dca29d039a5c9ea13e622e496ca61e85c34. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/whisper-large-v3-turbo-FP8-dynamic/blob/e72a6dca29d039a5c9ea13e622e496ca61e85c34/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 6 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-whisper-large-v3-turbo-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-whisper-large-v3-turbo-quantized-w4a16",
    "name": "RedHatAI/whisper-large-v3-turbo-quantized.w4a16 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/whisper-large-v3-turbo-quantized.w4a16/blob/2009b1e9931565204da231c92ee9dce46034efc8/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/whisper-large-v3-turbo-quantized.w4a16 model card, revision 2009b1e9931565204da231c92ee9dce46034efc8. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/whisper-large-v3-turbo-quantized.w4a16/blob/2009b1e9931565204da231c92ee9dce46034efc8/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 15 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-whisper-large-v3-turbo-quantized-w4a16-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-whisper-large-v3-turbo-quantized-w8a8",
    "name": "RedHatAI/whisper-large-v3-turbo-quantized.w8a8 model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/whisper-large-v3-turbo-quantized.w8a8/blob/95991fd06170e03c8fa55595958972e6cb76c441/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/whisper-large-v3-turbo-quantized.w8a8 model card, revision 95991fd06170e03c8fa55595958972e6cb76c441. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/whisper-large-v3-turbo-quantized.w8a8/blob/95991fd06170e03c8fa55595958972e6cb76c441/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 7 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-whisper-large-v3-turbo-quantized-w8a8-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-voxtral-mini-3b-2507-fp8-dynamic",
    "name": "RedHatAI/Voxtral-Mini-3B-2507-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Voxtral-Mini-3B-2507-FP8-dynamic/blob/b6694dd7dfc0d8c4bff962f04a39ffc98f50b2aa/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Voxtral-Mini-3B-2507-FP8-dynamic model card, revision b6694dd7dfc0d8c4bff962f04a39ffc98f50b2aa. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Voxtral-Mini-3B-2507-FP8-dynamic/blob/b6694dd7dfc0d8c4bff962f04a39ffc98f50b2aa/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 26 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-voxtral-mini-3b-2507-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "quant-redhatai-voxtral-small-24b-2507-fp8-dynamic",
    "name": "RedHatAI/Voxtral-Small-24B-2507-FP8-dynamic model card (quantized vs unquantized evaluation)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/RedHatAI/Voxtral-Small-24B-2507-FP8-dynamic/blob/a5dba05c66c6f06ec358dbf6531087a858ab141d/README.md",
    "checked": "2026-09-25",
    "attribution": "Red Hat AI (Neural Magic), RedHatAI/Voxtral-Small-24B-2507-FP8-dynamic model card, revision a5dba05c66c6f06ec358dbf6531087a858ab141d. Selected own evaluation cells (quantized checkpoint vs unquantized baseline) transcribed; percentages / 100; retention recomputed. No endorsement.",
    "match": [
      "huggingface.co/RedHatAI/Voxtral-Small-24B-2507-FP8-dynamic/blob/a5dba05c66c6f06ec358dbf6531087a858ab141d/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned README YAML metadata line 26 declares `license: apache-2.0`; no separate LICENSE file in the repository at this revision. Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction and distribution of Derivative Works subject to preserving the licence and marking changes. Only Red Hat's own quantized-vs-unquantized evaluation cells are used; vendor comparison tables are not.",
    "license_url": "/licenses/quant-redhatai-voxtral-small-24b-2507-fp8-dynamic-Apache-2.0.txt"
  },
  {
    "id": "speed-meta-muse-glimmer-30b",
    "name": "meta-models/Muse-Glimmer-30B model card (speed and benchmark tables)",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/meta-models/Muse-Glimmer-30B/blob/a4e59da52a7bc87ae7251dd5545c0dd437c44b68/LICENSE",
    "checked": "2026-10-09",
    "attribution": "Meta Superintelligence Lab, meta-models/Muse-Glimmer-30B model card, revision a4e59da52a7bc87ae7251dd5545c0dd437c44b68. Selected published speed cells and benchmark cells (including Meta's internal run of Gemma 4 31B) transcribed. No endorsement.",
    "match": [
      "huggingface.co/meta-models/Muse-Glimmer-30B/blob/a4e59da52a7bc87ae7251dd5545c0dd437c44b68/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository contains the full Apache License 2.0 as LICENSE and the README YAML declares `license: apache-2.0` (line 2); the card also states \"License: Apache 2.0\". Apache-2.0 sections 1, 2 and 4 cover documentation in Source form and permit reproduction/derivative works subject to preserving the licence and marking changes. USAGE_POLICY.md governs use of the model, not reuse of the card's measurements. RedHatAI/Muse-Glimmer-30B carries a byte-identical copy of this README; the first-party original is cited.",
    "license_url": "/licenses/speed-meta-muse-glimmer-30b-Apache-2.0.txt",
    "scope": " Benchmark cells only where Meta's linked methodology report says Meta ran them; cells it says were sourced from Artificial Analysis (τ³-Banking, Terminal-Bench 2.1, SciCode, MMMU-Pro, and any marked (AA)) are not used (removed 2026-10-09). Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "quant-qwen-hf-qwen3-0-6b-gptq-int8",
    "name": "Qwen/Qwen3-0.6B-GPTQ-Int8 model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-0.6B-GPTQ-Int8/blob/d3f20e7e71825fe57ad32705b396e9e156279107/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-0.6B-GPTQ-Int8 model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and GPTQ-int8 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-0.6B-GPTQ-Int8/blob/d3f20e7e71825fe57ad32705b396e9e156279107/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-0-6b-gptq-int8-Apache-2.0.txt"
  },
  {
    "id": "quant-qwen-hf-qwen3-1-7b-gptq-int8",
    "name": "Qwen/Qwen3-1.7B-GPTQ-Int8 model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-1.7B-GPTQ-Int8/blob/382fed3fa21d1a7e837b42a500564a0c4be4f060/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-1.7B-GPTQ-Int8 model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and GPTQ-int8 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-1.7B-GPTQ-Int8/blob/382fed3fa21d1a7e837b42a500564a0c4be4f060/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-1-7b-gptq-int8-Apache-2.0.txt"
  },
  {
    "id": "quant-qwen-hf-qwen3-4b-awq",
    "name": "Qwen/Qwen3-4B-AWQ model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-4B-AWQ/blob/74d4bd2bd4bff9cafc9345221320bffb08b406a3/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-4B-AWQ model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and AWQ-int4 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-4B-AWQ/blob/74d4bd2bd4bff9cafc9345221320bffb08b406a3/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-4b-awq-Apache-2.0.txt"
  },
  {
    "id": "quant-qwen-hf-qwen3-8b-awq",
    "name": "Qwen/Qwen3-8B-AWQ model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-8B-AWQ/blob/4da05a8edb55c6046cce958586c33b61da07bb79/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-8B-AWQ model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and AWQ-int4 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-8B-AWQ/blob/4da05a8edb55c6046cce958586c33b61da07bb79/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-8b-awq-Apache-2.0.txt"
  },
  {
    "id": "quant-qwen-hf-qwen3-14b-awq",
    "name": "Qwen/Qwen3-14B-AWQ model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-14B-AWQ/blob/31c69efc29464b6bb0aee1398b5a7b50a99340c3/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-14B-AWQ model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and AWQ-int4 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-14B-AWQ/blob/31c69efc29464b6bb0aee1398b5a7b50a99340c3/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-14b-awq-Apache-2.0.txt"
  },
  {
    "id": "quant-qwen-hf-qwen3-32b-awq",
    "name": "Qwen/Qwen3-32B-AWQ model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-32B-AWQ/blob/0499c3ac83fdef8810b907a23894ba91e95eddd8/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-32B-AWQ model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and AWQ-int4 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-32B-AWQ/blob/0499c3ac83fdef8810b907a23894ba91e95eddd8/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-32b-awq-Apache-2.0.txt"
  },
  {
    "id": "quant-qwen-hf-qwen3-30b-a3b-gptq-int4",
    "name": "Qwen/Qwen3-30B-A3B-GPTQ-Int4 model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-GPTQ-Int4/blob/9b534e4318b7ebc3c961a839f13eb18b1833f441/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-30B-A3B-GPTQ-Int4 model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and GPTQ-int4 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-30B-A3B-GPTQ-Int4/blob/9b534e4318b7ebc3c961a839f13eb18b1833f441/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-30b-a3b-gptq-int4-Apache-2.0.txt"
  },
  {
    "id": "quant-qwen-hf-qwen3-235b-a22b-gptq-int4",
    "name": "Qwen/Qwen3-235B-A22B-GPTQ-Int4 model card — quantized vs bf16 performance table",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-GPTQ-Int4/blob/dff138951f3898772fc0d795a4b5b30f417577f4/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Qwen Team, Alibaba Cloud. Qwen/Qwen3-235B-A22B-GPTQ-Int4 model card, Apache-2.0 (Copyright 2025 Alibaba Cloud). Selected bf16 and GPTQ-int4 table cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "huggingface.co/Qwen/Qwen3-235B-A22B-GPTQ-Int4/blob/dff138951f3898772fc0d795a4b5b30f417577f4/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Pinned repository root LICENSE is the Apache License, Version 2.0 ('Copyright 2025 Alibaba Cloud') and the README metadata declares 'license: apache-2.0'. Apache-2.0 s.1 defines Source form to include documentation source; s.2 grants the right to reproduce, prepare derivative works of, publicly display and distribute the Work; s.4 requires retaining the licence/notices and marking changes. Only the card's own quantized-vs-bf16 table cells are used.",
    "license_url": "/licenses/quant-qwen-hf-qwen3-235b-a22b-gptq-int4-Apache-2.0.txt"
  },
  {
    "id": "speed-llama-cpp-benches-nemotron",
    "name": "ggml-org/llama.cpp — committed Nemotron DGX Spark benchmark document",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/ggml-org/llama.cpp/blob/27b20ba8b1b85fc800b0a04caa8f1003ca64a2a3/LICENSE",
    "checked": "2026-09-25",
    "attribution": "The ggml authors. Selected published llama-bench tg measurements, identifiers normalized. No endorsement.",
    "match": [
      "github.com/ggml-org/llama.cpp/blob/27b20ba8b1b85fc800b0a04caa8f1003ca64a2a3/benches/nemotron/nemotron-dgx-spark.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices. Same pinned revision and LICENSE as the existing speed-llama-cpp-benches entry; could instead be added as a third match path of that entry.",
    "license_url": "/licenses/speed-llama-cpp-benches-nemotron-MIT.txt"
  },
  {
    "id": "speed-mlx-lm-benchmarks",
    "name": "ml-explore/mlx-lm — committed BENCHMARKS.md (speed and MMLU-Pro by precision)",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/ml-explore/mlx-lm/blob/87b7b583a697537aa68f47130b40884700b5f55f/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright © 2023 Apple Inc. (mlx-lm). Selected generation-speed and MMLU-Pro cells; percentages converted to fractions and retention ratios derived. No endorsement.",
    "match": [
      "github.com/ml-explore/mlx-lm/blob/87b7b583a697537aa68f47130b40884700b5f55f/mlx_lm/BENCHMARKS.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices. One entry covers both the speed rows and the MMLU-Pro retention findings from the same committed file.",
    "license_url": "/licenses/speed-mlx-lm-benchmarks-MIT.txt"
  },
  {
    "id": "speed-strix-halo-guide",
    "name": "hogeheer499-commits/strix-halo-guide — first-party Strix Halo benchmark data (data/benchmarks.csv and cited raw llama-bench CSVs)",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Copyright (c) 2026 Hoge-heer-499. Selected first-party (measured-local/historical-local) rows; community_* files excluded; identifiers normalized. No endorsement.",
    "match": [
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/benchmarks.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-03/qwen3.6-35b-a3b-ud-q4-k-m-b9010-r20-run2.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-03/rocm-b8460-current/qwen3-coder-30b-ud-q4-k-xl-b8460-hip-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-03/rocm-b8460-current/qwen3.6-35b-ud-q4-k-m-b8460-hip-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/gpt-oss-120b-local-attempt/gpt-oss-120b-mxfp4-b9049-vulkan-tg128-r3.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/gpt-oss-120b-local-attempt/gpt-oss-120b-mxfp4-b9049-vulkan-tg32-r3.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/latest-stack-rerun/clean-b9049-rerun/qwen3-coder-30b-b9049-clean-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/latest-stack-rerun/clean-b9049-rerun/qwen36-35b-b9049-clean-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/max-performance-campaign/benchmarks/gpt-oss-120b-long-context-vulkan/gpt-oss-120b-tg128-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/max-performance-campaign/benchmarks/qwen3-coder-top-confirm-r20/guide.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/max-performance-campaign/benchmarks/qwen36-top-confirm-r20/q4-0-ub2048.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-07/max-performance-campaign/benchmarks/qwen36-top-confirm-r20/q4-k-m-ub1024.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-16/break-97-24-strict-noise-settings/b9179-q4-k-s-r50.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-16/latest-stack-b9172/qwen3-next-confirm-r20/qwen3-next-80b-b9172-ub1024-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-16/lemonade-rocm-b1259-spotcheck/qwen3-next-80b-rocm-tg128-r3.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-16/qwen3-coder-break100-master/guide-r20/q4-k-s.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-26/latest-llamacpp-b9334/b9334-qwen3-coder-q4-k-s-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-26/latest-llamacpp-b9334/b9334-qwen3-coder-ud-q4-k-xl-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-27/latest-llamacpp-b9360/b9360-qwen3-coder-q4-k-s-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-27/latest-llamacpp-b9360/b9360-qwen3-coder-ud-q4-k-xl-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-05-31/latest-llamacpp-b9442-safe-clean/qwen3-coder-q4-k-s-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-01/latest-llamacpp-de6f727-safe-clean/qwen3-coder-q4-k-s-tg128-mmp0-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-02/modern-model-clean-followup/qwen3-coder-next-iq4xs-b9467-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-02/qwen3-30b-a3b-2507-direct-scout/qwen3-30b-2507-iq4xs-b9467-r50.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-03/qwen3-coder-iq4xs-direct-scout/pp512-tg128-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-03/qwen35-35b-a3b-iq4xs-direct-scout/pp512-tg128-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-05/latest-llamacpp-intdot-regression/lfm25-8b-a1b-q4km-b2016bf2-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-05/latest-llamacpp-intdot-regression/nemotron-3-nano-30b-a3b-iq4xs-b2016bf2-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-05/latest-llamacpp-intdot-regression/nemotron-3-super-120b-a12b-udiq4xs-b2016bf2-r3.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-05/latest-llamacpp-intdot-regression/qwen3-30b-2507-iq4xs-b2016bf2-r3.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-05/latest-llamacpp-intdot-regression/qwen3-coder-30b-udq4kxl-b2016bf2-p0-n128-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-07/latest-llamacpp-b9544-regression/lfm25-8b-a1b-q4km-b9544-vulkan-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-07/latest-llamacpp-b9544-regression/nemotron-super-120b-udiq4xs-b9544-vulkan-r3.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-07/latest-llamacpp-b9544-regression/qwen3-30b-2507-iq4xs-b9544-vulkan-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-07/latest-llamacpp-b9544-regression/qwen3-coder-ud-q4kxl-b9544-vulkan-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-07/qwen3-coder-q4ks-b9544-refresh/qwen3-coder-q4ks-b9544-p512-n128-r50.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-11/latest-llamacpp-ac4cddeb-vulkan-clean/gemma-4-12b-it-qat-udq4kxl-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-11/latest-llamacpp-ac4cddeb-vulkan-clean/gemma-4-26b-a4b-it-qat-udq4kxl-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-11/latest-llamacpp-ac4cddeb-vulkan-clean/lfm25-8b-a1b-q4km-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-11/latest-llamacpp-ac4cddeb-vulkan-clean/nemotron-3-super-120b-a12b-udiq4xs-r3.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-11/latest-llamacpp-ac4cddeb-vulkan-clean/qwen3-30b-a3b-2507-iq4xs-r10.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-11/latest-llamacpp-ac4cddeb-vulkan-clean/qwen3-coder-30b-q4ks-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-30/latest-llamacpp-b9851-vulkan-sentinel/gemma4-26b-a4b-udq4km-b9851-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-30/latest-llamacpp-b9851-vulkan-sentinel/qwen3-coder-q4ks-b9851-p512-n128-r50.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-06-30/latest-llamacpp-b9851-vulkan-sentinel/qwen3-coder-udq4kxl-b9851-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-02/latest-llamacpp-b9859-vulkan-sentinel/gemma4-26b-a4b-udq4km-b9859-faon-mmp0-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-02/latest-llamacpp-b9859-vulkan-sentinel/qwen3-coder-q4ks-b9859-faon-mmp0-p512-n128-r50.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-02/latest-llamacpp-b9859-vulkan-sentinel/qwen3-coder-udq4kxl-b9859-faon-mmp0-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-06/latest-llamacpp-b9888-vulkan-sentinel/qwen3-coder-q4ks-b9888-faon-mmp0-p512-n128-r50.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-06/latest-llamacpp-b9888-vulkan-sentinel/qwen3-coder-udq4kxl-b9888-faon-mmp0-r5.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-16/llamacpp-b10046-rocm-integrated-host-buffer/bench.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-16/nemotron-cascade2-iq4xs/llama-bench.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-07-18/gemma4-31b-qat-dflash-b10066/llama-bench.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-08-30/b10687-vulkan-sentinel/qwen3-coder-30b-udq4kxl-b10687-p512-n128-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-08-30/b10687-vulkan-sentinel/qwen3-next-80b-udq4kxl-b10687-p512-n128-r20.csv",
      "github.com/hogeheer499-commits/strix-halo-guide/blob/dac7bb99938d9c4248a79532bfc89c29b9e59795/data/raw/2026-08-30/qwen38-flash-next-scout/qwen38-flash-next-udiq4xs-b10687-p512-n128-r10.csv"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "MIT expressly includes associated documentation files and grants use, modification, publication and distribution; retain copyright and permission notices. README 'License' section: 'This project is licensed under the MIT License.' data/README.md separates first-party benchmarks.csv from community_*.csv files, which are not used.",
    "license_url": "/licenses/speed-strix-halo-guide-MIT.txt"
  },
  {
    "id": "speed-stepfun-step35-llamacpp",
    "name": "stepfun-ai/Step-3.5-Flash — committed llama.cpp deployment benchmark doc (llama.cpp/docs/step3.5-flash.md)",
    "kind": "first-party",
    "license": "Apache-2.0 AND MIT",
    "terms_url": "https://github.com/stepfun-ai/Step-3.5-Flash/blob/21d85a5f6c291f3f138da0bc09979af43345251a/LICENSE",
    "checked": "2026-09-25",
    "attribution": "StepFun (Step-3.5-Flash repository, Apache-2.0) within its vendored llama.cpp subtree (MIT, The ggml authors). Selected S_TG cells. No endorsement.",
    "match": [
      "github.com/stepfun-ai/Step-3.5-Flash/blob/21d85a5f6c291f3f138da0bc09979af43345251a/llama.cpp/docs/step3.5-flash.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Repository root LICENSE is Apache-2.0 (sections 1, 2 and 4 cover documentation and derivative works); the file sits in the vendored llama.cpp/ subtree whose llama.cpp/LICENSE is MIT and expressly covers associated documentation files. Both notices reproduced in full.",
    "license_url": "/licenses/speed-stepfun-step35-llamacpp-Apache-2.0-MIT.txt"
  },
  {
    "id": "speed-diffusers-paraattention-doc",
    "name": "huggingface/diffusers — committed ParaAttention optimization guide (docs/source/en/optimization/para_attn.md)",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/huggingface/diffusers/blob/bdc2bea37a36038c44452811610489ea30ede229/LICENSE",
    "checked": "2026-09-25",
    "attribution": "The HuggingFace Team and contributors; ParaAttention guide contributed in PR #10544. Selected single-GPU baseline/FP8 timings. No endorsement.",
    "match": [
      "github.com/huggingface/diffusers/blob/bdc2bea37a36038c44452811610489ea30ede229/docs/source/en/optimization/para_attn.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Apache-2.0 sections 1, 2 and 4 cover documentation source, reproduction and derivative works; preserve license/notices and identify modifications. No root NOTICE file in the pinned tree.",
    "license_url": "/licenses/speed-diffusers-paraattention-doc-Apache-2.0.txt"
  },
  {
    "id": "speed-lhl-strix-halo-testing",
    "name": "lhl/strix-halo-testing — committed llm-bench per-model result tables",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/lhl/strix-halo-testing/blob/5ee14b5b0ecfbd16b7fd0aa41a91581c32b33b61/LICENSE",
    "checked": "2026-09-25",
    "attribution": "lhl (Leonard Lin) strix-halo-testing. Selected best tg128 rows; identifiers normalized. No endorsement.",
    "match": [
      "github.com/lhl/strix-halo-testing/blob/5ee14b5b0ecfbd16b7fd0aa41a91581c32b33b61/llm-bench/GLM-4.5-Air-UD-Q4_K_XL/README.md",
      "github.com/lhl/strix-halo-testing/blob/5ee14b5b0ecfbd16b7fd0aa41a91581c32b33b61/llm-bench/Llama-4-Scout-17B-16E-Instruct-UD-Q4_K_XL/README.md",
      "github.com/lhl/strix-halo-testing/blob/5ee14b5b0ecfbd16b7fd0aa41a91581c32b33b61/llm-bench/Qwen3-235B-A22B-Instruct-2507-UD-Q3_K_XL/README.md",
      "github.com/lhl/strix-halo-testing/blob/5ee14b5b0ecfbd16b7fd0aa41a91581c32b33b61/llm-bench/gpt-oss-120b-F16/README.md",
      "github.com/lhl/strix-halo-testing/blob/5ee14b5b0ecfbd16b7fd0aa41a91581c32b33b61/llm-bench/gpt-oss-20b-F16/README.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Apache-2.0 sections 1, 2 and 4 cover documentation source, reproduction and derivative works; preserve license/notices and identify modifications. No NOTICE file in the pinned tree.",
    "license_url": "/licenses/speed-lhl-strix-halo-testing-Apache-2.0.txt"
  },
  {
    "id": "quant-auto-round-docs",
    "name": "intel/auto-round — committed accuracy documents (GGUF, AutoScheme, MXFP4/NVFP4)",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/intel/auto-round/blob/a4f01f8584a69f4af8376239a5fa51331a6747be/LICENSE",
    "checked": "2026-09-25",
    "attribution": "Intel Corporation and auto-round contributors. Selected BF16/F16-baselined accuracy cells; retention ratios derived. No endorsement.",
    "match": [
      "github.com/intel/auto-round/blob/a4f01f8584a69f4af8376239a5fa51331a6747be/docs/auto_scheme_acc.md",
      "github.com/intel/auto-round/blob/a4f01f8584a69f4af8376239a5fa51331a6747be/docs/gguf_alg_ext_acc.md",
      "github.com/intel/auto-round/blob/a4f01f8584a69f4af8376239a5fa51331a6747be/docs/opt_rtn.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Apache-2.0 sections 1, 2 and 4 cover documentation source, reproduction and derivative works; preserve license/notices and identify modifications. No NOTICE file in the pinned tree.",
    "license_url": "/licenses/quant-auto-round-docs-Apache-2.0.txt"
  },
  {
    "id": "quant-tensorrt-model-optimizer-windows",
    "name": "NVIDIA/TensorRT-Model-Optimizer — committed Windows benchmark reference (examples/windows/Benchmark.md)",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/ed7e87953c1a3ed6a21c8bf6eac90565fa6b7a36/LICENSE",
    "checked": "2026-09-25",
    "attribution": "NVIDIA Corporation, TensorRT Model Optimizer. Selected MMLU FP16/INT4 cells; percentages / 100 and retention derived. No endorsement.",
    "match": [
      "github.com/NVIDIA/TensorRT-Model-Optimizer/blob/ed7e87953c1a3ed6a21c8bf6eac90565fa6b7a36/examples/windows/Benchmark.md"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Apache-2.0 sections 1, 2 and 4 cover documentation source, reproduction and derivative works; preserve license/notices and identify modifications. Root LICENSE also carries third-party code notices, reproduced in full.",
    "license_url": "/licenses/quant-tensorrt-model-optimizer-windows-Apache-2.0.txt"
  },
  {
    "id": "speed-siliconbench-v1",
    "name": "SiliconBench: Speed, Memory, and Fidelity for LLM Serving on Unified-Memory Desktops (arXiv:2609.19169v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2609.19169v1",
    "checked": "2026-09-25",
    "attribution": "Ranran Haoran Zhang, Aysa Xuemo Fan, David Munhá Correia, Alex Cheema, Rui Zhang, SiliconBench: Speed, Memory, and Fidelity for LLM Serving on Unified-Memory Desktops, arXiv:2609.19169v1, 12 Sep 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2609.19169v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2609.19169v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/speed-siliconbench-v1-CC-BY-4.0.txt"
  },
  {
    "id": "speed-basert-v1",
    "name": "BaseRT: Best-in-Class LLM Inference on Apple Silicon via Native Metal (arXiv:2607.00501v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2607.00501v1",
    "checked": "2026-09-25",
    "attribution": "Prabod Rathnayaka, Fabian Waschkowski, Lukas Wesemann, BaseRT: Best-in-Class LLM Inference on Apple Silicon via Native Metal, arXiv:2607.00501v1, 1 Jul 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2607.00501v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2607.00501v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/speed-basert-v1-CC-BY-4.0.txt"
  },
  {
    "id": "speed-basert-m5-v1",
    "name": "BaseRT: Advancing Best-in-Class LLM Inference with Apple M5 Neural Accelerators (arXiv:2607.19438v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2607.19438v1",
    "checked": "2026-09-25",
    "attribution": "Fabian Waschkowski, Prabod Rathnayaka, Lukas Wesemann, BaseRT: Advancing Best-in-Class LLM Inference with Apple M5 Neural Accelerators, arXiv:2607.19438v1, 21 Jul 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2607.19438v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2607.19438v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/speed-basert-m5-v1-CC-BY-4.0.txt"
  },
  {
    "id": "speed-silicon-showdown-v2",
    "name": "Silicon Showdown: Performance, Efficiency, and Ecosystem Barriers in Consumer-Grade LLM Inference (arXiv:2605.00519v2)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2605.00519v2",
    "checked": "2026-09-25",
    "attribution": "Abdurrahman Javat, Allan Kazakov, Silicon Showdown: Performance, Efficiency, and Ecosystem Barriers in Consumer-Grade LLM Inference, arXiv:2605.00519v2, 4 May 2026 (v2), CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2605.00519v2"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2605.00519v2 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/speed-silicon-showdown-v2-CC-BY-4.0.txt"
  },
  {
    "id": "speed-greenbench-v1",
    "name": "GreenBench: Benchmarking Energy Efficiency and Carbon Footprint of Open-Source LLM Inference on Apple Silicon (arXiv:2608.28667v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2608.28667v1",
    "checked": "2026-09-25",
    "attribution": "Rajeswari Kannan, Raj Firke, Shreya Bengle, Srushti Deshmukh, GreenBench: Benchmarking Energy Efficiency and Carbon Footprint of Open-Source LLM Inference on Apple Silicon, arXiv:2608.28667v1, 24 Aug 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2608.28667v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2608.28667v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/speed-greenbench-v1-CC-BY-4.0.txt"
  },
  {
    "id": "quant-mt-ptq-v1",
    "name": "The Uneven Impact of Post-Training Quantization in Machine Translation (arXiv:2508.20893v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2508.20893v1",
    "checked": "2026-09-25",
    "attribution": "Benjamin Marie, Atsushi Fujita, The Uneven Impact of Post-Training Quantization in Machine Translation, arXiv:2508.20893v1, 28 Aug 2025, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2508.20893v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2508.20893v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-mt-ptq-v1-CC-BY-4.0.txt"
  },
  {
    "id": "quant-codegen-ptq-v1",
    "name": "Quantize with Confidence? An Empirical Study of Quantization for Code Generation (arXiv:2607.14181v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2607.14181v1",
    "checked": "2026-09-25",
    "attribution": "Saima Afrin, Md. Zahidul Haque, Antonio Mastropaolo, Quantize with Confidence? An Empirical Study of Quantization for Code Generation, arXiv:2607.14181v1, 15 Jul 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2607.14181v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2607.14181v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-codegen-ptq-v1-CC-BY-4.0.txt"
  },
  {
    "id": "quant-lowbit-reasoning-v1",
    "name": "Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery (arXiv:2606.02011v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2606.02011v1",
    "checked": "2026-09-25",
    "attribution": "Ekaterina Alimaskina, Darya Rudas, Denis Shveykin, Gleb Molodtsov, Pavel Vasiliev, Aleksandr Beznosikov, Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery, arXiv:2606.02011v1, 1 Jun 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2606.02011v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2606.02011v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-lowbit-reasoning-v1-CC-BY-4.0.txt"
  },
  {
    "id": "quant-bangla-ptq-v1",
    "name": "Quantization Effects on Bangla Language Understanding in Large Language Models: A Systematic Evaluation (arXiv:2608.24615v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2608.24615v1",
    "checked": "2026-09-25",
    "attribution": "Ismail Hossain, Nafi Ullah Shafin, Mohammad Abdullah Al Mumin, Quantization Effects on Bangla Language Understanding in Large Language Models: A Systematic Evaluation, arXiv:2608.24615v1, 25 Aug 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2608.24615v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2608.24615v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-bangla-ptq-v1-CC-BY-4.0.txt"
  },
  {
    "id": "quant-consumer-blackwell-v1",
    "name": "Private LLM Inference on Consumer Blackwell GPUs: A Practical Guide for Cost-Effective Local Deployment (arXiv:2601.09527v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2601.09527v1",
    "checked": "2026-09-25",
    "attribution": "Jonathan Knoop, Hendrik Holtmann, Private LLM Inference on Consumer Blackwell GPUs: A Practical Guide for Cost-Effective Local Deployment, arXiv:2601.09527v1, 14 Jan 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2601.09527v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2601.09527v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-consumer-blackwell-v1-CC-BY-4.0.txt"
  },
  {
    "id": "quant-reasoning-compression-v3",
    "name": "When Reasoning Meets Compression: Understanding the Effects of LLMs Compression on Large Reasoning Models (arXiv:2504.02010v3)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2504.02010v3",
    "checked": "2026-09-25",
    "attribution": "Nan Zhang, Eugene Kwek, Yusen Zhang, Ngoc-Hieu Nguyen, Prasenjit Mitra, Rui Zhang, When Reasoning Meets Compression: Understanding the Effects of LLMs Compression on Large Reasoning Models, arXiv:2504.02010v3, 2 Mar 2026 (v3), CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2504.02010v3"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2504.02010v3 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-reasoning-compression-v3-CC-BY-4.0.txt"
  },
  {
    "id": "quant-nvfp4-qad-v3",
    "name": "Quantization-Aware Distillation for NVFP4 Inference Accuracy Recovery (arXiv:2601.20088v3)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2601.20088v3",
    "checked": "2026-09-25",
    "attribution": "Meng Xin, Sweta Priyadarshi, Jingyu Xin, et al. (NVIDIA), Quantization-Aware Distillation for NVFP4 Inference Accuracy Recovery, arXiv:2601.20088v3, 3 Mar 2026 (v3), CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2601.20088v3"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2601.20088v3 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-nvfp4-qad-v3-CC-BY-4.0.txt"
  },
  {
    "id": "quant-clinical-gptq-v1",
    "name": "The Effect of Quantization on Clinical Benchmarks: Accuracy and Safety Across Model Families (arXiv:2609.22216v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2609.22216v1",
    "checked": "2026-09-25",
    "attribution": "Leonard Twagirayezu, Prasenjit Mitra, The Effect of Quantization on Clinical Benchmarks: Accuracy and Safety Across Model Families, arXiv:2609.22216v1, 2 Sep 2026, CC BY 4.0. Selected table cells transcribed into JSON; percentages divided by 100 and ratios/means computed where stated in each row's transformation; model and hardware labels normalized. No endorsement.",
    "match": [
      "arxiv.org/html/2609.22216v1"
    ],
    "verdict": "allowed",
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The versioned abs page https://arxiv.org/abs/2609.22216v1 links the licence http://creativecommons.org/licenses/by/4.0/ and the HTML rendering states 'License: CC BY 4.0'. The grant is the paper's own, not inferred from model or code licences; only the authors' own measurements are selected.",
    "license_url": "/licenses/quant-clinical-gptq-v1-CC-BY-4.0.txt"
  },
  {
    "id": "whatmodel-api-speed",
    "name": "whatmodel.sh API speed measurements",
    "kind": "first-party",
    "license": "CC-BY-SA-4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://whatmodel.sh/method#speed",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "match": [
      "whatmodel.sh/data/api-speed.json"
    ],
    "icon": "whatmodel.sh",
    "attribution": "Measured by whatmodel.sh (Merlinsbeard Pty Ltd) with scripts/measure-api.ts; raw runs and settings in data/api-speed.json.",
    "quote": "Our own measurements of our own API requests; published under the site's data licence (CC BY-SA 4.0)."
  },
  {
    "id": "whatmodel-bench",
    "name": "whatmodel.sh community bench submissions",
    "kind": "community",
    "license": "CC0-1.0",
    "license_url": "https://creativecommons.org/publicdomain/zero/1.0/",
    "terms_url": "https://whatmodel.sh/bench",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "match": [
      "whatmodel.sh/data/bench.json"
    ],
    "icon": "whatmodel.sh",
    "attribution": "Measurements submitted through whatmodel.sh/bench and reviewed before publication; each submitter released them under CC0 1.0.",
    "quote": "Submitters tick 'I ran this on my machine, and I release these measurements under CC0 1.0' before submitting (app/bench/review.tsx)."
  },
  {
    "id": "first-party-fish-audio",
    "name": "Fish Audio official pricing documentation",
    "match": [
      "docs.fish.audio/developer-guide/models-pricing/pricing-and-rate-limits"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.fish.audio/developer-guide/models-pricing/pricing-and-rate-limits",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "fish.audio",
    "attribution": "Fish Audio official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "visual-genexam",
    "name": "GenExam: A Multidisciplinary Text-to-Image Exam (arXiv v5)",
    "match": [
      "arxiv.org/html/2509.14232v5"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-genexam.txt",
    "attribution": "Zhaokai Wang, Penghao Yin, Xiangyu Zhao, Changyao Tian, Yu Qiao, Wenhai Wang, Jifeng Dai, Gen Luo. GenExam: A Multidisciplinary Text-to-Image Exam, arXiv:2509.14232v5 (14 May 2026), CC BY 4.0. Extracted the Table 2 relaxed Overall score for selected models; 0-100 divided by 100; model labels mapped only to exactly identified models. No values estimated or rerun. No endorsement.",
    "quote": "arXiv abs page for 2509.14232v5 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2509.14232v5\".",
    "scope": "Benchmark authors' own evaluation in Table 2 (Sec. 4.1: \"We select 17 representative models for comparison ... We use the default T2I configuration for each of the models for inference\"; Sec. 4: GPT-5 evaluator, reasoning effort low). Seedream 4.0 row included: it is absent from v4 Table 2 and appears in v5 without a citation, so it is the authors' own v5 run although the Sec. 4.1 inventory text was not updated. Excluded: Janus-Pro (size not stated), BAGEL non-thinking (thinking row kept). Strict scores retained only in notes; appendix evaluator ablations not used.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-t2i-reasonbench",
    "name": "T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026)",
    "match": [
      "aclanthology.org/2026.findings-acl.433.pdf"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-t2i-reasonbench.txt",
    "attribution": "Kaiyue Sun, Rongyao Fang, Chengqi Duan, Xian Liu, Aoxue Li, Xihui Liu. T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation. Findings of the Association for Computational Linguistics: ACL 2026, pages 8919-8944. doi:10.18653/v1/2026.findings-acl.433. Licensed CC BY 4.0 via the ACL Anthology. Extracted the Table 2 Overall column for selected models; values divided by 100; model labels mapped only to the stated model. No values estimated or rerun. No endorsement.",
    "quote": "ACL Anthology site footer (https://aclanthology.org/2026.findings-acl.433/): \"Materials published in or after 2016 are licensed on a Creative Commons Attribution 4.0 International License.\" The arXiv preprint 2508.17472v1 is arXiv non-exclusive and is NOT used.",
    "scope": "Only the benchmark authors' own Table 2 evaluation (\"We evaluate 16 state-of-the-art T2I models ... We use the default settings for all T2I models in image generation\"). Pinned by Anthology ID (no revision v2 exists; the unversioned PDF is the version of record) plus PDF sha256 354ea7e432f4bc3c60b406374575c2f9280db02c32e2b02d74d68b9b5df824e3. Table 3 (LLM-rewritten prompts) and Table 4 (text-module) experiments not used.",
    "icon": "aclanthology.org"
  },
  {
    "id": "visual-tiif-bench",
    "name": "TIIF-Bench: How Does Your T2I Model Follow Your Instructions? (arXiv v3)",
    "match": [
      "arxiv.org/html/2506.02161v3"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-tiif-bench.txt",
    "attribution": "Xinyu Wei, Jinrui Zhang, Zeqing Wang, Hongyang Wei, Zhen Guo, Bairui Li, Lei Zhang. TIIF-Bench: How Does Your T2I Model Follow Your Instructions?, arXiv:2506.02161v3 (12 Jul 2026), CC BY 4.0. Extracted Table 2 short-prompt Overall and Table 3 GenAI-Bench Overall for selected models; 0-100 scores divided by 100; model labels mapped only to exactly identified models. No values estimated or rerun. No endorsement.",
    "quote": "arXiv abs page for 2506.02161v3 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/ (v1-v3 all CC BY 4.0). GitHub repo A113N-W3I/TIIF-Bench has no licence and is NOT used.",
    "scope": "Benchmark authors' own runs: Sec. 4.1 \"We benchmark TIIF-Bench on a diverse suite of widely used closed-source and leading open-source text-to-image models\" (Table 2) and Sec. 4.2 \"all models are evaluated on the complete prompt set using the benchmark's official evaluation protocol\" (Table 3, GenAI-Bench). Rows with unspecified variants (SD 3, SD 3.5, SANA, PixArt, Janus-Pro, Show-o, Lumina-mGPT, Hunyuan-DiT) excluded. CompBench++ Table 4 not used (two conflicting protocols, AVG is not an official overall). Long-prompt Overall retained only in notes.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-wise",
    "name": "WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)",
    "match": [
      "arxiv.org/html/2503.07265v4"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-SA-4.0",
    "terms_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-wise.txt",
    "attribution": "Yuwei Niu, Munan Ning, Mengren Zheng, Weiyang Jin, Bin Lin, Peng Jin, Jiaqi Liao, Chaoran Feng, Fanqing Meng, Kunpeng Ning, Bin Zhu, Li Yuan. WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation, arXiv:2503.07265v4 (2 Jun 2026), CC BY-SA 4.0. Extracted the Table 1 Overall WiScore for selected models; model labels mapped only to the stated checkpoint. No values estimated or rerun. Adapted rows must be shared under CC BY-SA 4.0. No endorsement.",
    "quote": "arXiv abs page for 2503.07265v4 links \"Rights to this article\" to http://creativecommons.org/licenses/by-sa/4.0/ (v1-v4 all CC BY-SA 4.0). The GitHub repo PKU-YuanGroup/WISE has no licence file and its leadboard.md is NOT used.",
    "scope": "Only the benchmark authors' own Table 1 evaluation (Sec. 4.1: \"We evaluate 20 T2I models ... we use the official default configurations of each model and fix the random seed\"; judge GPT-4o-2024-05-13). Table 2 (rewritten prompts), appendix judge/weight/seed sensitivity tables and the Table 11 supplementary evaluation (judge not stated) are not used. ShareAlike: publish in the separate CC BY-SA tranche.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-scivbench",
    "name": "Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains",
    "match": [
      "arxiv.org/html/2608.09873v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-scivbench.txt",
    "attribution": "Diandian Zhang, Tingyu Song, Lin Fu, Zheyuan Yang, Yilun Zhao (Tongji University, Yale University, Zhejiang University, UCAS). Sci-VBench, arXiv:2608.09873v1, CC BY 4.0. Extracted Table 4 Human Eval Avg. values; 1–5 ratings divided by 5; model labels mapped only to the exact API id / HF repository in Table 6. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2608.09873v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version.",
    "scope": "Only the authors' own expert human ratings of videos they generated for all 16 models (Table 4, Human Eval Avg.). Kling-2.6 omitted (API mode/tier not stated). Automatic MLLM-judge columns and Full Avg. not imported (same generated videos; would duplicate the experiment).",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-fepbench",
    "name": "Faithful, Enriched, and Precise: Benchmarking Natural-Science Illustration Generation by T2I models (FEPBench)",
    "match": [
      "arxiv.org/html/2606.05949v2"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC0-1.0",
    "terms_url": "https://creativecommons.org/publicdomain/zero/1.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-fepbench.txt",
    "attribution": "Yifan Chang, Jiaxin Ai, Jianwen Sun, Yuandong Pu, Siqi Luo, Liangliang Zhao, Yuchen Ren, Minghao Liu, Yunfei Yu, Yu Qiao, Kaipeng Zhang, Yihao Liu (Shanghai AI Laboratory, Shanghai Innovation Institute, ZODA et al.). FEPBench, arXiv:2606.05949v2, dedicated to the public domain under CC0 1.0 (credit given as courtesy). Extracted Table 2 IF (instruction faithfulness) means; model labels mapped only to the stated release. No values estimated or rerun.",
    "quote": "arXiv abstract page for 2606.05949v2 links http://creativecommons.org/publicdomain/zero/1.0/ (CC0 1.0 public domain dedication) for this version.",
    "scope": "Authors' own generation and GPT-5.4 atom-level scoring of nine T2I models (Table 2). Only the IF mean is imported; RE and SP means are other dimensions of the same experiment and are not imported. 'Seedream 5.0' is mapped to Seedream 5.0 Lite because the row's reference [3] is 'Seedream 5.0 Lite' (https://seed.bytedance.com/en/seedream5_0_lite, accessed 2026-05-06); 'Ovis-Image' is mapped to Ovis-Image-7B (reference [26], the only public checkpoint). The benchmark's reference figures are NC-licensed journal material and are not used.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-edit-compass",
    "name": "Edit-Compass & EditReward-Compass: A Unified Benchmark for Image Editing and Reward Modeling",
    "match": [
      "arxiv.org/html/2605.13062v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-edit-compass.txt",
    "attribution": "Xuehai Bai, Yang Shi, Yi-Fan Zhang, Xuanyu Zhu, Yuran Wang, Yifan Dai, Xinyu Liu, Yiyan Ji, Xiaoling Gu, Yuanxing Zhang (HDU, PKU, Kling Team, CASIA). Extracted the English-instruction Overall (AVG) cells of Table 3 for selected models; the 1–5 judge scale is divided by 5. No values estimated or rerun. Source: https://arxiv.org/html/2605.13062v1. No endorsement; provided as-is.",
    "quote": "arXiv abs page for 2605.13062v1 links License https://creativecommons.org/licenses/by/4.0/ ; HTML header: \"License: CC BY 4.0 arXiv:2605.13062v1 [cs.CV] 13 May 2026\". Permission covers the authors' own measurements in this article only; no permission inferred for generated images, dataset inputs or external websites.",
    "scope": "Benchmark authors' own evaluation of 29 editing models (Appendix D: 'We evaluate 29 mainstream image editing models'; closed models 'through their official API services'); Gemini-3.1-Pro judge. English Table 3 Overall only; Chinese Table 4 and per-dimension columns not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-geditbench-v2",
    "name": "GEditBench v2: A Human-Aligned Benchmark for General Image Editing",
    "match": [
      "arxiv.org/html/2603.28547v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-geditbench-v2.txt",
    "attribution": "Zhangqi Jiang, Zheng Sun, Xianfang Zeng, Yufeng Yang, Xuanyang Zhang, Yongliang Wu, Wei Cheng, Gang Yu, Xu Yang, Bihan Wen (Nanyang Technological University, StepFun, Southeast University). Extracted the Overall Elo column of Table 3 (OpenEdit leaderboard) for selected models. No values estimated or rerun. Source: https://arxiv.org/html/2603.28547v1. No endorsement; provided as-is.",
    "quote": "arXiv abs page for 2603.28547v1 links License https://creativecommons.org/licenses/by/4.0/ ; HTML header: \"License: CC BY 4.0 arXiv:2603.28547v1 [cs.CV] 30 Mar 2026\". Permission covers the authors' own measurements only. The table's 'Arena' column is cited from Artificial Analysis and is NOT used.",
    "scope": "Authors' own pairwise evaluation of 16 editing models on 1,200 GEditBench v2 queries (Sec 5.2 'We evaluate 16 representative image editing models ... Default hyperparameter settings are used'). Overall Elo only; per-dimension Elo and the Artificial Analysis Arena column excluded. Step1X-Edit-v1p2 is a StepFun model and StepFun staff are co-authors, so that row is lab.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-unicbench",
    "name": "UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits",
    "match": [
      "arxiv.org/html/2512.02790v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-unicbench.txt",
    "attribution": "Keming Ye, Zhipeng Huang, Canmiao Fu, Qingyang Liu, Jiani Cai, Zheqi Lv, Chen Li, Jing Lyu, Zhou Zhao, Shengyu Zhang (Zhejiang University, WeChat Vision Tencent, SJTU, Xinjiang University). Extracted UnicBench Overall-EN cells of Table 4 for selected models; the 0–10 scale is divided by 10. No values estimated or rerun. Source: https://arxiv.org/html/2512.02790v1. No endorsement; provided as-is.",
    "quote": "arXiv abs page for 2512.02790v1 links License https://creativecommons.org/licenses/by/4.0/ ; HTML header: \"License: CC BY 4.0 arXiv:2512.02790v1 [cs.CV] 01 Dec 2025\". Permission covers the authors' own measurements only; no permission inferred for images, the UnicEdit-10M dataset contents or external sites.",
    "scope": "Authors' own UnicBench runs (Sec 6.1: 'we test some representative models on UnicBench ... All assessments are conducted by gpt-4.1-2025-04-14'). Overall-EN only; CN columns and per-metric columns not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-tecci",
    "name": "TECCI: Tricky Edits of Collected and Curated Images",
    "match": [
      "arxiv.org/html/2606.01213v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-tecci.txt",
    "attribution": "Aishwarya Agrawal, Roy Hirsch, Yasumasa Onoe, Sherry Ben, Jason Baldridge (Google Research, Google DeepMind). Extracted the IRCS+GGIS human Overall success-rate cells of Table 2; percentages divided by 100. No values estimated or rerun. Source: https://arxiv.org/html/2606.01213v1. No endorsement; provided as-is.",
    "quote": "arXiv abs page for 2606.01213v1 links License https://creativecommons.org/licenses/by/4.0/ ; HTML header: \"License: CC BY 4.0 arXiv:2606.01213v1 [cs.CV] 31 May 2026\". Permission covers the authors' own human-study measurements only; no permission inferred for images or the arena.ai leaderboard mentioned in the footnote.",
    "scope": "Authors' own human evaluation of five proprietary editors obtained 'directly through their respective official APIs with default hyper-parameters' (Sec 4.1). Google authors evaluating Google models -> Nano Banana rows are lab. Table 2 combined IRCS+GGIS Overall only; per-criterion rates, subset rates, SxS Elo (Table 3) and autorater Table 4 not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-skyreels-v2",
    "name": "SkyReels-V2: Infinite-length Film Generative Model",
    "match": [
      "arxiv.org/html/2504.13074v3"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-skyreels-v2.txt",
    "attribution": "SkyReels Team (Skywork AI): Guibin Chen, Dixuan Lin, Jiangping Yang, et al. SkyReels-V2, arXiv:2504.13074v3, CC BY 4.0. Extracted baseline rows of Tables 5, 6 and 7; 1–5 human ratings divided by 5, percentages divided by 100; model labels mapped only to the stated release/mode. No values estimated or rerun. Source: https://arxiv.org/html/2504.13074v3. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2504.13074v3 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version (v1–v3 all CC BY 4.0).",
    "scope": "Skywork's own human evaluation on SkyReels-Bench (T2V Table 5, I2V Table 7) and own VBench 1.0 long-prompt runs (Table 6) of baseline models. Own SkyReels-V2 rows omitted (checkpoint size/resolution not stated); Wan2.1 I2V row omitted (480P vs 720P checkpoint not stated).",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-videophy-readme",
    "name": "VideoPhy / VideoPhy-2 official repository leaderboards",
    "match": [
      "github.com/Hritikbansal/videophy/blob/01d366c979e4a82875852adbf427ccc036e7373a/VIDEOPHY2/README.md",
      "github.com/Hritikbansal/videophy/blob/01d366c979e4a82875852adbf427ccc036e7373a/README.md"
    ],
    "exact": true,
    "kind": "leaderboard",
    "license": "MIT",
    "terms_url": "https://github.com/Hritikbansal/videophy/blob/01d366c979e4a82875852adbf427ccc036e7373a/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-videophy-readme.txt",
    "attribution": "VideoPhy and VideoPhy-2 authors (Hritik Bansal et al.), github.com/Hritikbansal/videophy at commit 01d366c979e4a82875852adbf427ccc036e7373a, MIT License, Copyright (c) 2024 Hritikbansal. Extracted leaderboard cells printed in the repository README files; percentages divided by 100; model labels mapped to the exact linked checkpoints. No values estimated or rerun. No endorsement.",
    "quote": "MIT License ... Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies ...",
    "scope": "Only the benchmark authors' own leaderboard cells printed in the MIT-licensed README files at this commit (VideoPhy-2 human leaderboard 'All' column; VideoPhy-1 automatic leaderboard 'Avg.' column; VideoPhy-1 human leaderboard 'SA=1, PC=1' joint column). Papers (arXiv non-exclusive) and HF datasets are not the licence basis. Sora (60-prompt UI subset) and models without a catalogue identity omitted.",
    "icon": "github.com"
  },
  {
    "id": "visual-v-reasonbench",
    "name": "V-ReasonBench: Toward Unified Reasoning Benchmark Suite for Video Generation Models",
    "match": [
      "arxiv.org/html/2511.16668v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-v-reasonbench.txt",
    "attribution": "Yang Luo, Xuanlei Zhao, Baijiong Lin, Lingting Zhu, Liyao Tang, Yuqi Liu, Ying-Cong Chen, Shengju Qian, Xin Wang, Yang You (NUS, HKUST(GZ), HKU, USYD, CUHK, LIGHTSPEED). V-ReasonBench, arXiv:2511.16668v1, CC BY 4.0. Extracted Table 2 Average pass@5; percentages divided by 100; model labels mapped only to the stated release/tier. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2511.16668v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version.",
    "scope": "Benchmark authors' own generation (five videos per prompt, default parameters) and answer-verified scoring of six commercial video models; Table 2 Average only. Vidu-Q2 omitted (tier not stated).",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-vwg-bench",
    "name": "From Evaluation to Enhancement: Benchmarking and Improving Think-with-Video Reasoning for Video Generative Models (VWG-Bench)",
    "match": [
      "arxiv.org/html/2609.11242v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-vwg-bench.txt",
    "attribution": "Meng Luo, Yicheng Liu, Jiahao Wang, Yuanxing Zhang, Xin Tao, Pengfei Wan, Kun Gai, Hao Fei (NUS, Nanjing University, Kling Team / Kuaishou Technology). arXiv:2609.11242v1, CC BY 4.0. Extracted VWG-Bench Overall (Table 2 and Appendix Table 3); 1–5 scores divided by 5; model labels mapped only to the stated release/tier. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2609.11242v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version.",
    "scope": "Authors' own baseline diagnostic without prompt extension (Table 2) plus the appendix Seedance 2.0 row under the same protocol (identical Veo3.1 row). Wan2.2 omitted (open checkpoint variant not stated). Vid-PRE-enhanced and prompt-extension rows not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-worldjen",
    "name": "WorldJen: An End-to-End Multi-Dimensional Benchmark for Generative Video Models",
    "match": [
      "arxiv.org/html/2605.03475v2"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-worldjen.txt",
    "attribution": "Karthik Inbasekar, Guy Rom, Omer Shlomovits (moonmath.ai). WorldJen, arXiv:2605.03475v2, CC BY 4.0. Extracted Table 1 human Bradley-Terry ratings; model labels mapped only to the stated fal.ai endpoint or local checkpoint. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2605.03475v2 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version (v1 and v2 both CC BY 4.0).",
    "scope": "Authors' own generation (fal.ai endpoints March 2026; Wan 2.1 1.3B locally) and own human pairwise study (2,696 votes, 7 annotators). Table 1 human BT only; the VLM BT leaderboard (Table 3) scores the same 300 videos and is not imported separately.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-t2vphysbench",
    "name": "T2VPhysBench: A First-Principles Benchmark for Physical Consistency in Text-to-Video Generation",
    "match": [
      "arxiv.org/html/2505.00337v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-t2vphysbench.txt",
    "attribution": "Xuyang Guo, Jiayan Huo, Zhenmei Shi, Zhao Song, Jiahao Zhang, Jiale Zhao. T2VPhysBench, arXiv:2505.00337v1, CC BY 4.0. Extracted Table 2 Avg. Score for models whose release is identified in Table 1; model labels mapped only to the stated release. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2505.00337v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version.",
    "scope": "Authors' own fully manual evaluation (three annotators per video) of ten T2V models; only Wan 2.1 (14B, open), Sora (2024), Mochi-1 and Pika 2.2 imported. Kling, Dreamina, Hailuo, LTX Video, Qingying and SD Video omitted (version/checkpoint not identified).",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-qcbench",
    "name": "QC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing)",
    "match": [
      "arxiv.org/html/2608.24112v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-qcbench.txt",
    "attribution": "Shaoan Zhao, Fang Zhao, Xueqiang Guo, Xinpei Su, Huanlin Gao, Qiang Hui, Ting Lu, Fuyuan Shi, Chao Tan, Bikun Yang, Kai Wang, Shiguo Lian (China Unicom Data Science & AI Research Institute / Unicom Data Intelligence). arXiv:2608.24112v1, CC BY 4.0. Extracted Table 2 English HCQ Total; percentages divided by 100; model labels mapped only to the stated checkpoint. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2608.24112v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version.",
    "scope": "Authors' own generation (official defaults, prompt enhancement disabled) and HCQ scoring of 13 open T2I models; Table 2 English Total only. Chinese track not imported (related track of the same experiment). HiDream-I1 (variant unstated) and Lens (not in catalogue) omitted.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-imagetime",
    "name": "Can Image Models Imagine Time? ImageTime: A Novel Benchmark for Probing Visual World Modeling Through Spatiotemporal Consistency",
    "match": [
      "arxiv.org/html/2606.10620v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-imagetime.txt",
    "attribution": "Xinrui Wu, Lichen Huang. ImageTime, arXiv:2606.10620v1, CC BY 4.0. Extracted Table 5 Overall mean (0–10 judge scale) divided by 10; model labels mapped only to the cited release. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2606.10620v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version.",
    "scope": "Authors' own prompt-only generation of 750 four-panel motion sheets per model and GPT-5.5 structured-judge scoring (Table 5 Overall mean). Author affiliations are not stated in the paper; none is attributed to an evaluated model's maker.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-gebench",
    "name": "GEBench: Benchmarking Image Generation Models as GUI Environments",
    "match": [
      "arxiv.org/html/2602.09007v2"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "license_url": "/licenses/visual-gebench.txt",
    "attribution": "Haodong Li, Jingwei Wu, Quan Sun, Guopeng Li, Juanxi Tian, Huanyu Zhang, Yanlin Lai, Ruichuan An, Hongbo Peng, Yuhong Dai, Chenxi Li, Chunmei Qing, Jia Wang, Ziyang Meng, Zheng Ge, Xiangyu Zhang, Daxin Jiang (StepFun, SCUT, PKU, Tsinghua, CASIA, UChicago, NTU). arXiv:2602.09007v2, CC BY 4.0. Extracted Table 1 English GE-Score; divided by 100; model labels mapped only to the stated release. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2602.09007v2 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version.",
    "scope": "Authors' own generation with official default configurations and VLM-judge scoring of 12 models (Table 1 English GE-Score). No StepFun model is evaluated. Omitted: Wan 2.6 (wan2.6-t2i vs wan2.6-image unclear), Longcat-Image (image vs -Edit unclear), UniWorld-V2 (not in catalogue); Chinese subset not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-imagenworld",
    "name": "ImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World Tasks",
    "match": [
      "arxiv.org/html/2603.27862v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "license_url": "/licenses/visual-imagenworld.txt",
    "attribution": "Samin Mahdizadeh Sani, Max Ku, Nima Jamali, Matina Mahdizadeh Sani, Paria Khoshtab, Wei-Chieh Sun, Parnian Fazel, Zhi Rui Tam, Thomas Chong, Edisy Kin Wai Chan, Donald Wai Tong Tsang, Chiao-Wei Hsu, Ting Wai Lam, Ho Yin Sam Ng, Chiafeng Chu, Chak-Wing Mak, Keming Wu, Hiu Tung Wong, Yik Chun Ho, Chi Ruan, Zhuofeng Li, I-Sheng Fang, Shih-Ying Yeh, Ho Kei Cheng, Ping Nie, Wenhu Chen (University of Waterloo et al.). arXiv:2603.27862v1, CC BY 4.0. Extracted Table 3 human-annotated overall (TIG and TIE); model labels mapped only to the stated release. No values estimated or rerun. No endorsement; provided as-is.",
    "quote": "arXiv abstract page for 2603.27862v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0); HTML header reads 'License: CC BY 4.0 arXiv:2603.27862v1'.",
    "scope": "Authors' own generation (paper/official default configurations, seed 42 for open models, closed models via API) and three-annotator human ratings; Table 3 O_Human for the TIG and TIE tasks only. InstructPix2Pix TIE row added 2026-09-29 (Table 2 cites Brooks et al. 2023; single public checkpoint timbrooks/instruct-pix2pix). Omitted: Gemini 2.0 Flash (exp vs preview-image-generation not stated), Step1X-Edit (version unstated), Janus Pro (size unstated), Infinity (2B vs 8B unstated), IC-Edit (LoRA variant unstated), UNO (subject-driven model, not added); reference-image tasks and VLM columns not imported. Extended 2026-10-09: Scope extension if accepted: Table 3 TIG row 'UNO' (-> new uno); remove 'UNO (subject-driven model, not added)' from the omitted list.",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-deepseek",
    "name": "DeepSeek official pricing and model documentation",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/blob/b1c0b44b4369b597ad119a196caf79a9c40e141e/config.json",
      "huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B/blob/6a6f4aa4197940add57724a7707d069478df56b1/config.json",
      "huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B/blob/1df8507178afcc1bef68cd8c393f61a886323761/config.json",
      "github.com/deepseek-ai/DeepSeek-LLM/blob/6712a86bfb7dd25c73383c5ad2eb7a8db540258b/README.md",
      "huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct/blob/e434a23f91ba5b4923cf6c9d9a238eb4a08e3a11/README.md",
      "huggingface.co/deepseek-ai/DeepSeek-V2-Lite-Chat/blob/85864749cd611b4353ce1decdb286193298f64c7/README.md",
      "huggingface.co/deepseek-ai/deepseek-coder-1.3b-base/blob/c919139c3a9b4070729c8b2cca4847ab29ca8d94/README.md",
      "huggingface.co/deepseek-ai/deepseek-coder-1.3b-instruct/blob/e063262dac8366fc1f28a4da0ff3c50ea66259ca/README.md",
      "huggingface.co/deepseek-ai/deepseek-coder-33b-base/blob/45c85cadf3720ef3e85a492e24fd4b8c5d21d8ac/README.md",
      "huggingface.co/deepseek-ai/deepseek-coder-33b-instruct/blob/61dc97b922b13995e7f83b7c8397701dbf9cfd4c/README.md",
      "huggingface.co/deepseek-ai/deepseek-coder-6.7b-base/blob/ce2207a8bfef3ee92bd7dd4cc31c52cfa0046912/README.md",
      "huggingface.co/deepseek-ai/deepseek-coder-6.7b-instruct/blob/e5d64addd26a6a1db0f9b863abf6ee3141936807/README.md",
      "huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B/blob/ad9f0ae0864d7fbcd1cd905e3c6c5b069cc8b562/config.json",
      "huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B/blob/916b56a44061fd5cd7d6a8fb632557ed4f724f60/config.json",
      "huggingface.co/deepseek-ai/DeepSeek-V2.5/blob/c85b5ede86f2a598af339624cac5723861e557ed/config.json",
      "huggingface.co/deepseek-ai/DeepSeek-V3.1-Terminus/blob/19510d6dc61f79dbd925bd51ee8a9081c509a4b6/config.json",
      "huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/config.json"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/blob/b1c0b44b4369b597ad119a196caf79a9c40e141e/config.json",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "exact": true,
    "icon": "deepseek.com",
    "attribution": "DeepSeek official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-gradium",
    "name": "Gradium official pricing and model documentation",
    "match": [
      "gradium.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://gradium.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "gradium.ai",
    "attribution": "Gradium official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-ibm",
    "name": "IBM official pricing and model documentation",
    "match": [
      "huggingface.co/ibm-granite/granite-4.1-8b/blob/1504002f650e656a0a3789d99574df12e3e94ed0/README.md",
      "huggingface.co/ibm-granite/granite-3.0-2b-instruct/blob/5ad66c190631382717bd92d7b052adb1a7b669e7/README.md",
      "huggingface.co/ibm-granite/granite-3.0-8b-instruct/blob/5af56291a1c6ca9056df597aab0cff53edabddb0/README.md",
      "huggingface.co/ibm-granite/granite-3.1-2b-instruct/blob/bbc2aed595bd38bd770263dc3ab831db9794441d/README.md",
      "huggingface.co/ibm-granite/granite-3.1-8b-instruct/blob/4009206d5fc95d2e65a7b7633e159d6e97e25d35/README.md",
      "huggingface.co/ibm-granite/granite-3.3-2b-instruct/blob/707f574c62054322f6b5b04b6d075f0a8f05e0f0/README.md",
      "huggingface.co/ibm-granite/granite-4.0-1b/blob/6a7381ba1f54d684ff508d991aeb7dc580157103/README.md",
      "huggingface.co/ibm-granite/granite-4.0-350m/blob/bd8a1497065c0d6ba1ef19af6b0d2b14bacf71c2/README.md",
      "ibm.com/docs/SSLSRPV_2.1.x/wsj/analyze-data/fm-models-ibm.html",
      "huggingface.co/ibm-granite/granite-3.2-2b-instruct/blob/641593c3b25bec0b1efe9f0f7d7a67f7243f86a3/config.json",
      "huggingface.co/ibm-granite/granite-3.0-3b-a800m-instruct/blob/51ced9b4d6ecb3a6671e2e52c937bc12a4a470a5/README.md",
      "huggingface.co/ibm-granite/granite-3.0-1b-a400m-instruct/blob/ffec3c35bdfd97a06f0b4cd5fcc92cd9b1584445/README.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://huggingface.co/ibm-granite/granite-4.1-8b/blob/1504002f650e656a0a3789d99574df12e3e94ed0/README.md",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "ibm.com",
    "attribution": "IBM official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-krea",
    "name": "Krea official pricing and model documentation",
    "match": [
      "krea.ai/docs/api-reference/krea/krea-2-turbo.md",
      "krea.ai/docs/developers/krea-2/overview.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://www.krea.ai/docs/developers/krea-2/overview.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "krea.ai",
    "attribution": "Krea official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-moonshot",
    "name": "Moonshot AI (Kimi) official pricing and model documentation",
    "match": [
      "platform.kimi.ai/docs/pricing/chat.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://platform.kimi.ai/docs/pricing/chat.md",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "kimi.ai",
    "attribution": "Moonshot AI (Kimi) official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-novita",
    "name": "Novita AI official pricing and model documentation",
    "match": [
      "novita.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://novita.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-09-25",
    "exact": true,
    "icon": "novita.ai",
    "attribution": "Novita AI official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-openai",
    "name": "OpenAI official pricing and model documentation",
    "match": [
      "developers.openai.com/api/docs/models/gpt-5.5-pro",
      "developers.openai.com/api/docs/models/gpt-4",
      "developers.openai.com/api/docs/models/gpt-3.5-turbo",
      "developers.openai.com/api/docs/models/gpt-4-turbo",
      "developers.openai.com/api/docs/models/gpt-4o",
      "developers.openai.com/api/docs/models/gpt-4o-mini",
      "developers.openai.com/api/docs/models/o1",
      "developers.openai.com/api/docs/models/o3-mini",
      "developers.openai.com/api/docs/models/gpt-4.1-nano",
      "developers.openai.com/api/docs/models/gpt-5-pro",
      "developers.openai.com/api/docs/models/gpt-5.2-pro",
      "developers.openai.com/api/docs/models/gpt-5.4-pro",
      "cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf",
      "cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf",
      "developers.openai.com/api/docs/models/gpt-6.1-sol",
      "developers.openai.com/api/docs/pricing",
      "developers.openai.com/api/docs/models/chatgpt-image-latest",
      "developers.openai.com/api/docs/models/gpt-4o-search-preview"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://developers.openai.com/api/docs/models/gpt-5.5-pro",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "openai.com",
    "attribution": "OpenAI official pricing/docs, observed 2026-09-25 GPT-6.1 Sol facts observed 2026-10-06.",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-poolside",
    "name": "Poolside official pricing and model documentation",
    "match": [
      "huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/README.md",
      "huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/README.md",
      "huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/README.md",
      "huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/README.md"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/README.md",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "exact": true,
    "icon": "poolside.ai",
    "attribution": "Poolside official pricing/docs, observed 2026-09-25",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "matharena-arxiv-outputs-bysa",
    "name": "MathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/MathArena/arxivmath-1225_outputs/blob/6e427c4860254adc93ee95ebd37dc2b6c6028b4a/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivmath-0126_outputs/blob/306b8e8d1e91c40beb6b2b003adb89e7951bd204/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivmath-0226_outputs/blob/92e08f2fb94f5fd7fdbf02b335f5ce6ff9a589a8/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivmath-0326_outputs/blob/69bc5be7b2280c8521854515a981ed074aaa7b40/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivmath-0426_outputs/blob/31b49f173ea747a53fb613b8c8f24b6b4bdf6fcf/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivmath-0526_outputs/blob/85f1009458c2d2e1527d3e592d04cf40ee8b3a90/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivmath-0626_outputs/blob/f782eef32107837304e2ecd4e78b4cb60ee1e122/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/brokenarxiv-0226_outputs/blob/b1346bc6e4bd465b9c163debae81126b1140bfea/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/brokenarxiv-0326_outputs/blob/d719c23340e971fe956a2503bd157bfdcb67d146/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/brokenarxiv-0426_outputs/blob/34779f2c3c287fda88ea06b4c7885a1d10ae8b36/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/brokenarxiv-0526_outputs/blob/7ca7ebc694e921b56eb9b5968374918bf2a2b4e4/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/brokenarxiv-0626_outputs/blob/c4fac0ea2939a252ea72e9712fa49c639becb252/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivlean-0326_outputs/blob/47c34bc76c06596e0010c5c15a93c23ca9f082ed/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivmath-0826_outputs/blob/64731d80cf7dd2a83d842c4deb9739c81d129efa/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/brokenarxiv-0826_outputs/blob/7da551be2a80d7721bd6ee77ef14c8894fdef686/data/train-00000-of-00001.parquet",
      "huggingface.co/datasets/MathArena/arxivlean-0626_outputs/blob/4221f7241f21cf4d39a80b934b7ea76d53ba29ed/data/train-00000-of-00001.parquet"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "terms_url": "https://huggingface.co/datasets/MathArena/arxivmath-0626_outputs/blob/f782eef32107837304e2ecd4e78b4cb60ee1e122/README.md",
    "license_url": "/licenses/matharena-sharealike-NOTICE.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "MathArena (Jasper Dekoninck, Nikola Jovanović, Tim Gehrunger, Kári Rögnvaldsson, Ivo Petrov, Chenhao Sun, Martin Vechev; SRI Lab, ETH Zurich), model-output datasets on Hugging Face (huggingface.co/MathArena), CC BY-SA 4.0. Adapted by whatmodel.sh: per-attempt correctness aggregated to per-model mean scores (mean over problems of the per-problem mean over attempts), only complete non-harness runs, highest configured reasoning effort per model, catalogue IDs mapped. No problem statements, model answers or transcripts redistributed. No endorsement.",
    "quote": "Each pinned dataset card declares YAML 'license: cc-by-sa-4.0' and states under 'Licensing Information': 'This dataset is licensed under the Attribution-ShareAlike 4.0 International (CC BY-SA 4.0). Please abide by the license when using the provided data.'",
    "scope": "Only the *_outputs datasets listed in match (ArXivMath Dec 2025-Aug 2026, BrokenArXiv Feb-Aug 2026, ArXivLean Mar and Jun 2026), each CC BY-SA 4.0 on its own card. MathArena's competition datasets (AIME, HMMT, USAMO, BRUMO, SMT, CMIMC, Kangaroo, Putnam, IMO/IMC, Apex, Project Euler and their _outputs) are CC BY-NC-SA 4.0 and are excluded. The MIT licence of the eth-sri/matharena GitHub harness is not relied on for the data; configs there are cited only for run settings.",
    "adapter_license": "CC-BY-SA-4.0",
    "adapter_license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "notice_file": "/licenses/matharena-sharealike-NOTICE.txt",
    "adapted_data_file": "/data/llm-new-bysa.json",
    "disclaimer": "Licensed material is provided as-is and as-available, without warranties; see CC BY-SA 4.0 section 5.",
    "publication_requirement": "Retain this tranche, attribution, changes, licence and provenance when displaying or serving its rows.",
    "icon": "huggingface.co"
  },
  {
    "id": "eqbench3-canonical-results",
    "name": "EQ-Bench 3 canonical leaderboard results",
    "kind": "dataset",
    "match": [
      "github.com/EQ-bench/eqbench3/blob/affc5a000bbdb965b807e09ca7875f8530fa8f56/data/canonical_leaderboard_elo_results.json.gz"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://github.com/EQ-bench/eqbench3/blob/affc5a000bbdb965b807e09ca7875f8530fa8f56/LICENSE",
    "license_url": "/licenses/eqbench3-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Samuel J. Paech, EQ-Bench 3 (github.com/EQ-bench/eqbench3), MIT, Copyright (c) 2025 Sam Paech. Selected ratings from the canonical leaderboard file; catalogue IDs mapped. No endorsement. Full licence in the linked notice.",
    "quote": "Root LICENSE: MIT License, Copyright (c) 2025 Sam Paech; README section 9: 'License: This project is open-sourced under an MIT license'. The canonical results file is committed in the same repository and README describes it as 'the official leaderboard data from models included on eqbench.com'; no nested licence override. The eqbench.com website itself is not used.",
    "icon": "github.com"
  },
  {
    "id": "galileo-agent-leaderboard",
    "name": "Galileo Agent Leaderboard (v1 and v2 results)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/galileo-ai/agent-leaderboard/blob/8158b75ae542c2e847ff6d517a0b66caa704b19a/results_v1.csv",
      "huggingface.co/spaces/galileo-ai/agent-leaderboard/blob/8158b75ae542c2e847ff6d517a0b66caa704b19a/results_v2.csv"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/galileo-ai/agent-leaderboard/blob/8158b75ae542c2e847ff6d517a0b66caa704b19a/README.md",
    "license_url": "/licenses/galileo-agent-leaderboard-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Galileo (galileo.ai), Agent Leaderboard, Hugging Face Space galileo-ai/agent-leaderboard, Apache-2.0. Selected rows from results_v1.csv and results_v2.csv; catalogue IDs mapped. No endorsement.",
    "quote": "Space card README.md at the pinned revision declares 'license: apache-2.0'; results_v1.csv and results_v2.csv are committed in the same Space repository with no other licence statement. Companion GitHub repo rungalileo/agent-leaderboard is MIT.",
    "icon": "huggingface.co"
  },
  {
    "id": "openai-simple-evals-readme",
    "name": "OpenAI simple-evals README benchmark results (OpenAI's own models)",
    "kind": "dataset",
    "match": [
      "github.com/openai/simple-evals/blob/652c89d0ca9df547706735883097e9537d40dc47/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://github.com/openai/simple-evals/blob/652c89d0ca9df547706735883097e9537d40dc47/LICENSE",
    "license_url": "/licenses/openai-simple-evals-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "OpenAI, simple-evals (github.com/openai/simple-evals), MIT, Copyright (c) 2024 OpenAI. Selected README table cells for OpenAI models; percentages converted to fractions; catalogue IDs mapped. Lab-reported. No endorsement.",
    "quote": "Root LICENSE: MIT License, Copyright (c) 2024 OpenAI ('the Software and associated documentation files'). The README states the repository exists 'so we can be transparent about the accuracy numbers we're publishing alongside our latest models'. Only OpenAI-model rows are used; the 'Other Models (Reported)' rows are copied vendor numbers and excluded.",
    "icon": "github.com"
  },
  {
    "id": "arena-hard-auto-readme-v2",
    "name": "Arena-Hard-Auto v2.0-Preview leaderboard (repository README)",
    "kind": "leaderboard",
    "match": [
      "github.com/lmarena/arena-hard-auto/blob/196f6b826783b3da7310e361a805fa36f0be83f3/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/lmarena/arena-hard-auto/blob/196f6b826783b3da7310e361a805fa36f0be83f3/LICENSE",
    "license_url": "/licenses/arena-hard-auto-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "LMArena / Arena-Hard-Auto authors (Tianle Li et al.), github.com/lmarena/arena-hard-auto, Apache-2.0. Selected README leaderboard rows; percentages converted to fractions; catalogue IDs mapped. No endorsement.",
    "quote": "Root LICENSE in the same pinned repository is the Apache License 2.0; the leaderboard tables are part of the repository README (documentation) with no separate licence. The lmarena.ai / arena.ai websites are not used.",
    "icon": "github.com"
  },
  {
    "id": "scicode-readme-leaderboard",
    "name": "SciCode leaderboard (repository README)",
    "kind": "leaderboard",
    "match": [
      "github.com/scicode-bench/SciCode/blob/e3158ea011d4235245a547460d3688d7ccbf9900/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/scicode-bench/SciCode/blob/e3158ea011d4235245a547460d3688d7ccbf9900/LICENSE",
    "license_url": "/licenses/scicode-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "SciCode authors (Minyang Tian et al.), github.com/scicode-bench/SciCode, Apache-2.0. Selected README rows; percentages converted to fractions. No endorsement.",
    "quote": "Root LICENSE in the same pinned repository is the Apache License 2.0; the leaderboard table is in the repository README with no separate licence. The scicode-bench.github.io site (no licence) is not used.",
    "icon": "github.com"
  },
  {
    "id": "judgemark-v2-results",
    "name": "Judgemark v2 results (EQ-Bench)",
    "kind": "dataset",
    "match": [
      "github.com/EQ-bench/Judgemark-v2/blob/2388448a0401c4d6ed5502ff18953adba944a255/results/scores.csv"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://github.com/EQ-bench/Judgemark-v2/blob/2388448a0401c4d6ed5502ff18953adba944a255/LICENSE",
    "license_url": "/licenses/judgemark-v2-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Samuel J. Paech, Judgemark v2 (github.com/EQ-bench/Judgemark-v2), MIT. Selected rows of results/scores.csv; scores divided by 100. No endorsement.",
    "quote": "Root LICENSE in the same pinned repository is the MIT License; results/scores.csv is committed there with no nested licence.",
    "icon": "github.com"
  },
  {
    "id": "arcprize-model-baseline-outputs",
    "name": "ARC Prize Foundation model_baseline outputs on ARC-AGI-1/2 public evaluation sets",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/blob/3e9c9d1a8402aff82356815c106cecaea65cb7d9/claude-opus-4-5-20251101-thinking-32k/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/blob/3e9c9d1a8402aff82356815c106cecaea65cb7d9/claude-opus-4-6-thinking-120K-max/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/blob/3e9c9d1a8402aff82356815c106cecaea65cb7d9/claude-opus-4-8-high/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/blob/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-1-2025-11-13-thinking-high/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/blob/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-2-2025-12-11-thinking-xhigh/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/deepseek-v3.2",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/glm-5",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/kimi-k2.5",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/minimax-m2.5",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/claude-opus-4-5-20251101-thinking-64k/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/claude-opus-4-6-thinking-120K-high/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/claude-opus-4-8-high/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gemini-3-flash-preview-thinking-high/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-5-1-2025-11-13-thinking-high/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-5-2-2025-12-11-thinking-medium/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/deepseek-v3.2",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/glm-5",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/kimi-k2.5",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/minimax-m2.5",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/blob/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-2-pro-2025-12-11-high/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/claude-3-haiku",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-4-mini-high",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-4-nano-high",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-5-2-pro-2025-12-11-medium/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/claude-3-haiku",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-5-4-mini-low",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-5-4-nano-medium",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-4o",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-4o",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/blob/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gemini-3-flash-preview-thinking-minimal/results.json",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/QwQ-32B-Fireworks",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/claude-haiku-4-5-20251001-thinking-32k",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/claude-sonnet-4-5-20250929-thinking-32k",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-4-1-2025-04-14",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-4-1-mini-2025-04-14",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-4-1-nano-2025-04-14",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-4-5-2025-02-27",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-4-high",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-4-pro-xhigh",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/gpt-5-pro-2025-10-06",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/grok-4-fast-reasoning",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/grok-4.20-beta-0309b-reasoning",
      "huggingface.co/datasets/arcprize/arc_agi_v1_public_eval/tree/3e9c9d1a8402aff82356815c106cecaea65cb7d9/qwen3-235b-a22b-instruct-2507",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/claude-haiku-4-5-20251001-thinking-32k",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/claude-sonnet-4-5-20250929-thinking-32k",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-4-1-2025-04-14",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-4-1-mini-2025-04-14",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-4-1-nano-2025-04-14",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-5-4-high",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/gpt-5-pro-2025-10-06",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/grok-4-fast-reasoning",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/grok-4.20-beta-0309b-reasoning",
      "huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/tree/026789c1c12a4c34580a32e84dcaf5630d7e8f31/qwen3-235b-a22b-instruct-2507"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/datasets/arcprize/arc_agi_v2_public_eval/blob/026789c1c12a4c34580a32e84dcaf5630d7e8f31/README.md",
    "license_url": "/licenses/arcprize-model-baseline-outputs-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "attribution": "ARC Prize Foundation, model outputs from github.com/arcprizeorg/model_baseline published as Hugging Face datasets arcprize/arc_agi_v1_public_eval and arcprize/arc_agi_v2_public_eval, MIT. Selected model runs; scores divided by task count or recomputed from the published per-attempt correctness flags; catalogue IDs mapped. No endorsement.",
    "quote": "Each dataset card declares YAML 'license: mit' and describes the data as 'the output of models tested with https://github.com/arcprizeorg/model_baseline'; that harness repository's LICENSE.md is 'MIT License, Copyright (c) 2025 ARC Prize Foundation'. The arcprize.org leaderboard website is not used.",
    "icon": "huggingface.co"
  },
  {
    "id": "vector-inspect-evals-leaderboard",
    "name": "Vector Institute Evaluation Leaderboard (Inspect Evals runs)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/vector-institute/eval-leaderboard/blob/f348507a79f56dae8624b525574cb326359a20c8/data/results.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/vector-institute/eval-leaderboard/blob/f348507a79f56dae8624b525574cb326359a20c8/README.md",
    "license_url": "/licenses/vector-inspect-evals-leaderboard-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Vector Institute, Evaluation Leaderboard (huggingface.co/spaces/vector-institute/eval-leaderboard), results run with Inspect Evals. Apache-2.0 (Space card). Selected published numerical results; model IDs mapped; no endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: apache-2.0`. src/about.py: 'All evaluations presented on this leaderboard are run using Inspect Evals ... we provide Inspect Logs for every benchmark run.' Results file data/results.json is committed in the same Space repository; no other licence or terms found.",
    "icon": "huggingface.co"
  },
  {
    "id": "bridge-medical-leaderboard",
    "name": "BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/YLab-Open/BRIDGE-Medical-Leaderboard/blob/579abc27d8659fce3ca357ee9108d4e279fbeb38/leaderboards/Zero-Shot_leaderboard.json"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/YLab-Open/BRIDGE-Medical-Leaderboard/blob/579abc27d8659fce3ca357ee9108d4e279fbeb38/README.md",
    "license_url": "/licenses/bridge-medical-leaderboard-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Jiageng Wu, Bowen Gu, Kevin Xie, Jie Yang et al., BRIDGE Medical Leaderboard (YLab-Open, Harvard Medical School / Brigham and Women's Hospital). MIT (Space card). Selected published zero-shot averages; percentages converted to fractions; model IDs mapped; no endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: mit`. docs.md: 'we construct this leaderboard of LLM in clinical text understanding by systematically evaluating 111 state-of-the-art LLMs' (maintained by the team of Prof. Jie Yang and Prof. Kueiyu Joshua Lin). No contrary terms in the Space.",
    "icon": "huggingface.co"
  },
  {
    "id": "roblox-llm-leaderboard",
    "name": "Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/TorpedoSoftware/Roblox-LLM-Leaderboard/blob/3fcd29933bf7276ce5470d9de8c235fe5684fc43/data/leaderboard.json",
      "huggingface.co/spaces/TorpedoSoftware/Roblox-LLM-Leaderboard/blob/676755cd82ed2886048e282f1a6106ddf6e5fea5/data/leaderboard.json"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/TorpedoSoftware/Roblox-LLM-Leaderboard/blob/3fcd29933bf7276ce5470d9de8c235fe5684fc43/README.md",
    "license_url": "/licenses/roblox-llm-leaderboard-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "Torpedo Software (boatbomber), Roblox LLM Leaderboard. MIT (Space card). Selected published raw counts converted to fractions; model IDs mapped; no endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: mit`. README: 'Every number on the board is transcribed from a run file's own summary; nothing is typed by hand. The build cross-checks each run against the harness cache ... every model must have been asked the same questions.'",
    "icon": "huggingface.co"
  },
  {
    "id": "webapp1k-leaderboard",
    "name": "WebApp1K models leaderboard (onekq-ai)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/onekq-ai/WebApp1K-models-leaderboard/blob/f05bc19c75983e09e6c3198ff13e28c8c9362ff0/results.csv",
      "huggingface.co/spaces/onekq-ai/WebApp1K-models-leaderboard/blob/f05bc19c75983e09e6c3198ff13e28c8c9362ff0/results_duo.csv"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/onekq-ai/WebApp1K-models-leaderboard/blob/f05bc19c75983e09e6c3198ff13e28c8c9362ff0/README.md",
    "license_url": "/licenses/webapp1k-leaderboard-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Yi Cui / onekq-ai, WebApp1K and WebApp1K-Duo leaderboard (per-problem run/success counts). MIT (Space card). pass@1 recomputed exactly as the Space's app.py does; model IDs mapped; no endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: mit`. The benchmark datasets onekq-ai/WebApp1K-React and WebApp1K-Duo-React are also MIT. The Space publishes per-problem Runs/Successes for each model (the maintainer's own generations; generations datasets are published separately).",
    "icon": "huggingface.co"
  },
  {
    "id": "mideind-icelandic-llm-leaderboard",
    "name": "Miðeind Icelandic LLM Leaderboard",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/mideind/icelandic-llm-leaderboard/blob/f8e42e15da1b985242bbb29eff139ec62405b53b/leaderboard.json",
      "huggingface.co/spaces/mideind/icelandic-llm-leaderboard/blob/7b12ee6799a4735154be85170c53d7266b2de4e6/leaderboard.json"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard/blob/f8e42e15da1b985242bbb29eff139ec62405b53b/README.md",
    "license_url": "/licenses/mideind-icelandic-llm-leaderboard-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "Miðeind ehf., Icelandic LLM Leaderboard (huggingface.co/spaces/mideind/icelandic-llm-leaderboard), CC BY 4.0. Adapted by whatmodel.sh: selected the published 'average' per model at its highest listed reasoning setting and mapped model IDs. No endorsement.",
    "quote": "README at the pinned revision: 'Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/): reuse freely with attribution.' Space card YAML: `license: cc-by-4.0`. Built and maintained by Miðeind.",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2605.09063v3",
    "name": "Soohak: A Mathematician-Curated Benchmark for Evaluating Research-level Math Capabilities of LLMs — authors' own model runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.09063v3"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2605.09063v3",
    "license_url": "/licenses/paper-2605.09063v3-CC-BY-SA-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Guijin Son, Seungone Kim, Catherine Arnett et al. (2026). Soohak: A Mathematician-Curated Benchmark for Evaluating Research-level Math Capabilities of LLMs. arXiv:2605.09063v3. https://arxiv.org/abs/2605.09063v3\nCC BY-SA 4.0: https://creativecommons.org/licenses/by-sa/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2605.09063v3 links the licence 'https://creativecommons.org/licenses/by-sa/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY-SA 4.0'.",
    "scope": "Table 2 (Avg@3 columns for SOOHAK-Mini and SOOHAK Challenge) with decoding settings from Table 4; all rows are the authors' own runs. Pass@3 and Refusal columns, Tables 5-6 (Pass@3 only) not imported.",
    "icon": "arxiv.org",
    "adapter_license": "CC-BY-SA-4.0",
    "publication_requirement": "Share-alike: retain attribution, change notice and licence when displaying or serving these rows; adapted rows must be offered under CC BY-SA 4.0.",
    "adapter_license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "adapted_data_file": "/data/llm-new-bysa.json"
  },
  {
    "id": "paper-2605.08498v1",
    "name": "MathConstraint: Automated Generation of Verified Combinatorial Reasoning Instances for LLMs — authors' own model runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.08498v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2605.08498v1",
    "license_url": "/licenses/paper-2605.08498v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Viresh Pati, Zhengyu Li, Piyush Jha et al. (2026). MathConstraint: Automated Generation of Verified Combinatorial Reasoning Instances for LLMs. arXiv:2605.08498v1. https://arxiv.org/abs/2605.08498v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2605.08498v1 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2 (no-tools accuracy on the full 329-instance release) with reasoning configs from Table 10; all rows are the authors' own runs. Tool-enabled (Table 4) and Easy-split (Table 3) results not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2608.30248v1",
    "name": "DSEffi-Bench: Demystifying Large Language Models' Capability in Efficient Data Science Code Generation — authors' own model runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2608.30248v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2608.30248v1",
    "license_url": "/licenses/paper-2608.30248v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Zhihao Gong, Junzhe Yu, Dong Huang et al. (2026). DSEffi-Bench: Demystifying Large Language Models' Capability in Efficient Data Science Code Generation. arXiv:2608.30248v1. https://arxiv.org/abs/2608.30248v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2608.30248v1 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2 Overall Pass column (greedy T=0.0), model identifiers from Appendix C.2; all rows are the authors' own runs. Efficiency (Beyond) columns not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2603.07316v1",
    "name": "FinSheet-Bench: From Simple Lookups to Complex Reasoning, Where LLMs Break on Financial Spreadsheets — authors' own model runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2603.07316v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2603.07316v1",
    "license_url": "/licenses/paper-2603.07316v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Jan Ravnik, Matjaž Ličen, Felix Bührmann et al. (2026). FinSheet-Bench: From Simple Lookups to Complex Reasoning, Where LLMs Break on Financial Spreadsheets. arXiv:2603.07316v1. https://arxiv.org/abs/2603.07316v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2603.07316v1 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 4 Acc. column (all listed catalogue models have 100% coverage); all rows are the authors' own runs.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2601.11868v1",
    "name": "Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces — benchmark authors' own agent-model trials",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2601.11868v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2601.11868v1",
    "license_url": "/licenses/paper-2601.11868v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Mike A. Merrill, Alexander G. Shaw, Nicholas Carlini et al. (2026). Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces. arXiv:2601.11868v1. https://arxiv.org/abs/2601.11868v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2601.11868v1 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2 rows with the Terminus 2 agent only (the authors' neutral scaffold); other scaffolds (Codex CLI, Claude Code, OpenHands, Mini-SWE-Agent, Gemini CLI) not imported to avoid mixing harnesses. All rows are the benchmark authors' own trials (>=5 runs per combination).",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.05405v2",
    "name": "Agents' Last Exam — authors' own fixed-harness model comparison",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.05405v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2606.05405v2",
    "license_url": "/licenses/paper-2606.05405v2-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Yiyou Sun, Xinyang Han, Weichen Zhang et al. (2026). Agents' Last Exam. arXiv:2606.05405v2. https://arxiv.org/abs/2606.05405v2\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2606.05405v2 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 1 panel \"LLM Model Comparison (fixed OpenClaw + GUI-as-Tool)\", Overall Pass Rate column; rows marked † (extra visual sub-agent) and harness-comparison panels not imported. All rows are the authors' own runs.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2605.01417v1",
    "name": "Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks — authors' own runs of 61 models",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.01417v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2605.01417v1",
    "license_url": "/licenses/paper-2605.01417v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Benjamin Warner, Ratna Sagari Grandhi, Max Kieffer et al. (2026). Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks. arXiv:2605.01417v1. https://arxiv.org/abs/2605.01417v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2605.01417v1 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 4 (all models by mean win rate on Medmarks-V) with sampling parameters from Table 8; all rows are the authors' own runs. Quantized duplicates (FP8/8-bit/4-bit rows) and lower-effort gpt-oss rows not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2602.13964v4",
    "name": "HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam — Qwen Team re-evaluation on revised Humanity's Last Exam",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2602.13964v4"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2602.13964v4",
    "license_url": "/licenses/paper-2602.13964v4-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Weiqi Zhai, Zhihai Wang, Jinghang Wang et al. (2026). HLE-Verified: A Systematic Verification and Structured Revision of Humanity's Last Exam. arXiv:2602.13964v4. https://arxiv.org/abs/2602.13964v4\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2602.13964v4 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2 \"Revised: HLE-Verified Full\" accuracy column (text-only subset) for models not made by the authors' organisation; raw-HLE and revised-subset columns not imported. Runs by the paper authors (Qwen Team, Alibaba Group); Qwen3-Max row not imported (authors' own model).",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2602.17831v2",
    "name": "The Token Games: Evaluating Language Model Reasoning with Puzzle Duels — authors' own tournament",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2602.17831v2"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2602.17831v2",
    "license_url": "/licenses/paper-2602.17831v2-CC-BY-SA-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Simon Henniger, Gabriel Poesia (2026). The Token Games: Evaluating Language Model Reasoning with Puzzle Duels. arXiv:2602.17831v2. https://arxiv.org/abs/2602.17831v2\nCC BY-SA 4.0: https://creativecommons.org/licenses/by-sa/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2602.17831v2 links the licence 'https://creativecommons.org/licenses/by-sa/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY-SA 4.0'.",
    "scope": "Table 1 TTG Elo column only (the authors' own 90-duel tournament). The HLE, ARC-AGI, SWE-Bench Pro, TextQuests and GPQA-D columns in the same table were taken from public leaderboards and are NOT imported.",
    "icon": "arxiv.org",
    "adapter_license": "CC-BY-SA-4.0",
    "publication_requirement": "Share-alike: retain attribution, change notice and licence when displaying or serving these rows; adapted rows must be offered under CC BY-SA 4.0.",
    "adapter_license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "adapted_data_file": "/data/llm-new-bysa.json"
  },
  {
    "id": "paper-2609.00654v1",
    "name": "SciTrue: Reliable Scientific Claim Validation with Frontier and Open Language Models at the NTCIR SciClaimEval Task — authors' own single-model dev evaluation",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.00654v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2609.00654v1",
    "license_url": "/licenses/paper-2609.00654v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Qiming Bao, Neşet Özkan Tan, Siyuan Wang et al. (2026). SciTrue: Reliable Scientific Claim Validation with Frontier and Open Language Models at the NTCIR SciClaimEval Task. arXiv:2609.00654v1. https://arxiv.org/abs/2609.00654v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2609.00654v1 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 3 rows marked Src=\"ours\" (run by the authors, refreshed 2026-08-02), T1 Pair column; rows marked \"official\" (organiser baselines) and T1 F1/T2 columns not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2605.22535v1",
    "name": "TerminalWorld: Benchmarking Agents on Real-World Terminal Tasks — authors' own Terminus-2 model comparison",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.22535v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2605.22535v1",
    "license_url": "/licenses/paper-2605.22535v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Zhaoyang Chu, Jiarui Hu, Xingyu Jiang et al. (2026). TerminalWorld: Benchmarking Agents on Real-World Terminal Tasks. arXiv:2605.22535v1. https://arxiv.org/abs/2605.22535v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2605.22535v1 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 1 Pass Rate (headline value, not the parenthesised error-excluded rate); rows for checkpoints that cannot be matched to catalogue identities not imported. All rows are the authors' own runs.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2604.18584v2",
    "name": "MathNet: a Global Multimodal Benchmark for Mathematical Reasoning and Retrieval — authors' own model runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2604.18584v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2604.18584v2",
    "license_url": "/licenses/paper-2604.18584v2-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "attribution": "Shaden Alshammari, Kevin Wen, Abrar Zainal et al. (2026). MathNet: a Global Multimodal Benchmark for Mathematical Reasoning and Retrieval. arXiv:2604.18584v2. https://arxiv.org/abs/2604.18584v2\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2604.18584v2 links the licence 'https://creativecommons.org/licenses/by/4.0/' in its 'View license' field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 3 Micro Avg. column on MathNet-Solve-Test; all rows are the authors' own runs. Ambiguous identities (ministral-3B, DeepSeek-R1) and non-catalogue models not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "inception-leaderboards-space",
    "name": "Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/inception42/Leaderboards/blob/bef41ddf1f6c39bb3a20ad13e518e28e4f2654fc/assets/results/aragen_v3_results.json",
      "huggingface.co/spaces/inception42/Leaderboards/blob/bef41ddf1f6c39bb3a20ad13e518e28e4f2654fc/assets/results/hindigen_v1_results.json",
      "huggingface.co/spaces/inception42/Leaderboards/blob/bef41ddf1f6c39bb3a20ad13e518e28e4f2654fc/assets/results/ifeval_results.jsonl"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/inception42/Leaderboards/blob/bef41ddf1f6c39bb3a20ad13e518e28e4f2654fc/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Inception (G42) / Ali El Filali et al., Multilingual Leaderboards Space (inception42/Leaderboards). Apache-2.0. Selected published numerical results; model IDs mapped and 0-1 scores kept as fractions. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: apache-2.0`. The result files are committed in the same Space repository; app.py states the evaluations are run by the maintainers on submitted models (private AraGen-v3 set, 3C3H judged by claude-sonnet-4-20250514)."
  },
  {
    "id": "speakleash-polish-eq-bench",
    "name": "SpeakLeash Polish EQ-Bench leaderboard",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/speakleash/polish_eq-bench/blob/ec5fcb7c7b9b9ee2fdc3e56651e59e2eb12c5765/benchmark_results.csv"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/speakleash/polish_eq-bench/blob/ec5fcb7c7b9b9ee2fdc3e56651e59e2eb12c5765/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "SpeakLeash.org Polish EQ-Bench Leaderboard (translation: Remigiusz Kinas; proofreading: Maria Filipkowska, Zuzanna Dabić; dataset: Kacper Milan; benchmark runs: Krzysztof Wróbel). Based on EQ-Bench by Samuel J. Paech. Apache-2.0. Selected rows; scores divided by 100. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: apache-2.0`. src/about.py: 'Running benchmark and leaderboard: Krzysztof Wróbel'."
  },
  {
    "id": "speakleash-cptub",
    "name": "SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/speakleash/cptu_bench/blob/7f981eb09374a686f238138b61e8ba8744251991/data.json"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/speakleash/cptu_bench/blob/7f981eb09374a686f238138b61e8ba8744251991/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "SpeakLeash.org CPTU-Bench Space. MIT. Selected rows; the Space's four-category Average reproduced and divided by 5. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: mit`. app.py documents the evaluation procedure (judge metamodel, 0-5 marks)."
  },
  {
    "id": "servicenow-browsergym-leaderboard",
    "name": "ServiceNow BrowserGym leaderboard (AgentLab GenericAgent runs)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4.1-Mini/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5-mini/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5-mini/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5-mini/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5-nano/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5-nano/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5-nano/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-5/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-oss-120b/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-oss-120b/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-oss-120b/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-oss-20b/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-oss-20b/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-oss-20b/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-405b/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-405b/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-405b/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-405b/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-70b/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-70b/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-70b/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Llama-3.1-70b/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-3.7-Sonnet/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-4-Sonnet/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-4-Sonnet/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-4-Sonnet/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o-mini/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o-mini/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o-mini/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o-mini/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-o1-mini/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-o1-mini/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-o1-mini/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-o1-mini/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-GPT-4o/workarena-l2.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-3.5-Sonnet/miniwob.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-3.5-Sonnet/webarena.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-3.5-Sonnet/workarena-l1.json",
      "huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/results/GenericAgent-Claude-3.5-Sonnet/workarena-l2.json"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/ServiceNow/browsergym-leaderboard/blob/294ebe1dc36ba83c6916c64c64fb870af3dc8cee/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "ServiceNow Research, BrowserGym Leaderboard Space; GenericAgent from AgentLab. MIT. Selected GenericAgent rows; percentages converted to fractions. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: mit`. Only GenericAgent-* folders (ServiceNow's own AgentLab agent, per each folder README) are imported; third-party agent submissions are excluded."
  },
  {
    "id": "open-pt-llm-leaderboard",
    "name": "Open Portuguese LLM Leaderboard (archived)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard/blob/788e0e7aa84b7dee80fc3aa3960beafb9e508ec9/data/leaderboard.csv.gz"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard/blob/788e0e7aa84b7dee80fc3aa3960beafb9e508ec9/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Eduardo A. S. Garcia and the Open PT-LLM Leaderboard (CEIA, Universidade Federal de Goiás). Apache-2.0. Selected rows of the archived static leaderboard; 0-100 averages divided by 100. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: apache-2.0`. frontend/src/content/about.md: models evaluated by the leaderboard on 9 Portuguese tasks with its lm-evaluation-harness fork; API models listed as 'external' in the same file."
  },
  {
    "id": "open-persian-llm-leaderboard",
    "name": "Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/opll-org/Open-Persian-LLM-Leaderboard/blob/0167a4aa7f90054c5d24c3ec8fd2ea89f0a60365/leaderboard_data.jsonl"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/opll-org/Open-Persian-LLM-Leaderboard/blob/0167a4aa7f90054c5d24c3ec8fd2ea89f0a60365/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Part DP AI and the Amirkabir University of Technology NLP Lab (Prof. Saeedeh Momtazi; Shahriar Shariati, Farhan Farsi, Shayan Bali), Open Persian LLM Leaderboard. Apache-2.0. Selected rows; the Space's 21-task Average Accuracy reproduced and divided by 100. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: apache-2.0`. utils.py about text: leaderboard developed by Part DP AI with AUT NLP Lab, evaluations on the LM Evaluation Harness; app.py computes 'Average Accuracy' as the mean of 21 task columns."
  },
  {
    "id": "mimic-cdm-leaderboard",
    "name": "MIMIC Clinical Decision Making leaderboard",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/MIMIC-CDM/leaderboard/blob/43e1ef313f61a726cfa30655284333ce7cfa641c/data/leaderboards.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/MIMIC-CDM/leaderboard/blob/43e1ef313f61a726cfa30655284333ce7cfa641c/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "MIMIC-CDM leaderboard (Hager et al., clinical decision making benchmark). Apache-2.0. Selected 16-bit rows; percentages divided by 100. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: apache-2.0`. Scores are bundled in data/leaderboards.json, synced from the maintainers' results-CDM (d7366a1b) and results-CDM-FI (d6bf5085) repositories."
  },
  {
    "id": "libra-leaderboard",
    "name": "ai-forever LIBRA leaderboard (Russian long-context benchmark)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/Qwen_Qwen3-30B-A3B-Instruct-2507.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/Qwen_Qwen3.5-0.8B.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/Qwen_Qwen3.5-122B-A10B.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/Qwen_Qwen3.5-2B.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/Qwen_Qwen3.5-35B-A3B.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/Qwen_Qwen3.5-4B.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/Qwen_Qwen3.5-9B.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/meta-llama_Llama-3.1-8B-Instruct.json",
      "huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/results/microsoft_Phi-3.5-mini-instruct.json"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard/blob/f5b4b2a59131f86a74af400036ff31b0aae74794/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "ai-forever (SberDevices) LIBRA leaderboard. MIT. Selected per-model results files; LIBRA Mini mean reproduced. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: mit`. docs/description.md recommends LIBRA Mini as the primary comparison suite."
  },
  {
    "id": "zai-rpc-bench-leaderboard",
    "name": "RPC-Bench leaderboard (Z.ai research paper comprehension)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/zai-org/RPC-Bench/blob/e9313f371c1646c5c711581490e01fcf165ebfab/data/leaderboard.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/zai-org/RPC-Bench/blob/e9313f371c1646c5c711581490e01fcf165ebfab/README.md",
    "license_url": "/licenses/hf-leaderboard-spaces-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Z.ai (zai-org), RPC-Bench leaderboard Space (paper arXiv:2601.14289). Apache-2.0. Selected text-input rows; percentages divided by 100. No endorsement.",
    "quote": "Space card YAML at the pinned revision: `license: apache-2.0`. data/leaderboard.json lists dataset rpc-bench-test-v1, eval.py@rpc-leaderboard-v1, judges gpt-5-2025-08-07 and gemini-2.5-pro for open QA and exact match for claim verification."
  },
  {
    "id": "multi-swe-bench-experiments",
    "name": "Multi-SWE-bench experiments (ByteDance Seed baseline runs)",
    "kind": "dataset",
    "match": [
      "github.com/multi-swe-bench/experiments/tree/6a7d5566f62fa76f4192302cf763051b98e4facc/evaluation"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/multi-swe-bench/experiments/blob/6a7d5566f62fa76f4192302cf763051b98e4facc/LICENSE",
    "license_url": "/licenses/multi-swe-bench-experiments-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "github.com",
    "attribution": "ByteDance Seed / Multi-SWE-bench team, multi-swe-bench/experiments. Apache-2.0. Resolved counts summed across seven language splits; model IDs mapped. No endorsement.",
    "quote": "Root LICENSE (Apache License 2.0) in the same pinned repository as the committed evaluation/<lang>/verified/*/results/results.json files; no nested licence override found. The pinned tree URL stands for the seven results.json files listed in each row's components."
  },
  {
    "id": "first-party-google-gemini-api",
    "name": "Google Gemini API official model documentation",
    "match": [
      "ai.google.dev/gemini-api/docs/models/gemini-3-flash-preview",
      "ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-flash-preview",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "exact": true,
    "icon": "ai.google.dev",
    "attribution": "Google Gemini API official model documentation, observed 2026-09-26",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-luma",
    "name": "Luma AI official pricing and model documentation",
    "match": [
      "docs.agents.lumalabs.ai/guides/pricing",
      "lumalabs.ai/api"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://docs.agents.lumalabs.ai/guides/pricing",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "exact": true,
    "icon": "lumalabs.ai",
    "attribution": "Luma AI official pricing/docs, observed 2026-09-26",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-siliconflow",
    "name": "SiliconFlow official pricing and model documentation",
    "match": [
      "siliconflow.com/models/glm-5v-turbo"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://www.siliconflow.com/models/glm-5v-turbo",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "exact": true,
    "icon": "siliconflow.com",
    "attribution": "SiliconFlow official pricing/docs, observed 2026-09-26",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-stability",
    "name": "Stability AI official pricing and model documentation",
    "match": [
      "platform.stability.ai/pricing",
      "huggingface.co/stabilityai/stable-code-3b/blob/5cee3fa2905e7a03c4a1b0bc02f39da7ceaa6cb5/README.md",
      "huggingface.co/stabilityai/stablelm-3b-4e1t/blob/fa4a6a92fca83c3b4223a3c9bf792887090ebfba/README.md",
      "huggingface.co/stabilityai/stablelm-base-alpha-3b/blob/819657db42861818efac86e8ca60f673fa912ef8/README.md",
      "huggingface.co/stabilityai/stablelm-base-alpha-7b/blob/1d43a3b40d7d708d8084d07dfae0c4e9fd95c04f/README.md",
      "huggingface.co/stabilityai/stablelm-base-alpha-3b-v2/blob/997d95943eee0697ba9e9cd2b981b9fad6962d9e/config.json",
      "huggingface.co/stabilityai/stablelm-base-alpha-7b-v2/blob/6d176503c1496b3c96a2bceb76a7d37695582618/config.json",
      "huggingface.co/stabilityai/stablelm-2-1_6b-chat/blob/f3fe67057c2789ae1bb1fe42b038da99840d4f13/config.json",
      "huggingface.co/stabilityai/stablelm-2-zephyr-1_6b/blob/2f275b1127d59fc31e4f7c7426d528768ada9ea4/config.json",
      "huggingface.co/stabilityai/stablelm-2-12b-chat/blob/b6b62cd451b84e848514c00fafa66d9ead9297c5/config.json",
      "huggingface.co/stabilityai/stablelm-zephyr-3b/blob/fe1fd5e36ccb8cb2cb1dc225ab6d2962692b9837/config.json"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://platform.stability.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "stability.ai",
    "attribution": "Stability AI official pricing/docs, observed 2026-09-26",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "card2-zai-org-glm-5",
    "name": "zai-org/GLM-5 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/zai-org--GLM-5.txt",
    "terms_url": "https://huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; no root LICENSE file; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "zai-org. zai-org/GLM-5 model card. Source: https://huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card2-zai-org-glm-4-7",
    "name": "zai-org/GLM-4.7 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/GLM-4.7/blob/602d01efcdd332c5238ca4bcede555defbe83eb7/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/zai-org--GLM-4.7.txt",
    "terms_url": "https://huggingface.co/zai-org/GLM-4.7/blob/602d01efcdd332c5238ca4bcede555defbe83eb7/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; no root LICENSE file; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "zai-org. zai-org/GLM-4.7 model card. Source: https://huggingface.co/zai-org/GLM-4.7/blob/602d01efcdd332c5238ca4bcede555defbe83eb7/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card2-zai-org-glm-4-7-flash",
    "name": "zai-org/GLM-4.7-Flash model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/GLM-4.7-Flash/blob/7dd20894a642a0aa287e9827cb1a1f7f91386b67/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/zai-org--GLM-4.7-Flash.txt",
    "terms_url": "https://huggingface.co/zai-org/GLM-4.7-Flash/blob/7dd20894a642a0aa287e9827cb1a1f7f91386b67/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; no root LICENSE file; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "zai-org. zai-org/GLM-4.7-Flash model card. Source: https://huggingface.co/zai-org/GLM-4.7-Flash/blob/7dd20894a642a0aa287e9827cb1a1f7f91386b67/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-stepfun-ai-step-3-5-flash",
    "name": "stepfun-ai/Step-3.5-Flash model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/stepfun-ai--Step-3.5-Flash.txt",
    "terms_url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "stepfun-ai. stepfun-ai/Step-3.5-Flash model card. Source: https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the starred ('*' = StepFun's reproduction under the same test conditions) competitor cells of the Detailed Benchmarks table, except xbench-DeepSearch rows."
  },
  {
    "id": "card2-deepseek-ai-deepseek-v4-1-flash",
    "name": "deepseek-ai/DeepSeek-V4.1-Flash model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/dba1be0a40aa45a94ad051997016db3960a90277/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/deepseek-ai--DeepSeek-V4.1-Flash.txt",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/dba1be0a40aa45a94ad051997016db3960a90277/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; root LICENSE begins 'MIT License'; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "deepseek-ai. deepseek-ai/DeepSeek-V4.1-Flash model card. Source: https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/dba1be0a40aa45a94ad051997016db3960a90277/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-deepseek-ai-deepseek-v4-pro-0813",
    "name": "deepseek-ai/DeepSeek-V4-Pro-0813 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/deepseek-ai--DeepSeek-V4-Pro-0813.txt",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; root LICENSE begins 'MIT License'; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "deepseek-ai. deepseek-ai/DeepSeek-V4-Pro-0813 model card. Source: https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-deepseek-ai-deepseek-v3-1",
    "name": "deepseek-ai/DeepSeek-V3.1 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V3.1/blob/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/deepseek-ai--DeepSeek-V3.1.txt",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1/blob/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; root LICENSE begins 'MIT License'; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "deepseek-ai. deepseek-ai/DeepSeek-V3.1 model card. Source: https://huggingface.co/deepseek-ai/DeepSeek-V3.1/blob/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card2-deepseek-ai-deepseek-v3-0324",
    "name": "deepseek-ai/DeepSeek-V3-0324 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V3-0324/blob/e9b33add76883f293d6bf61f6bd89b497e80e335/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/deepseek-ai--DeepSeek-V3-0324.txt",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/blob/e9b33add76883f293d6bf61f6bd89b497e80e335/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; root LICENSE begins 'MIT License'; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "deepseek-ai. deepseek-ai/DeepSeek-V3-0324 model card. Source: https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/blob/e9b33add76883f293d6bf61f6bd89b497e80e335/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-deepseek-ai-deepseek-v3-2",
    "name": "deepseek-ai/DeepSeek-V3.2 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/README.md",
      "huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/assets/paper.pdf"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/deepseek-ai--DeepSeek-V3.2.txt",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; root LICENSE begins 'MIT License'; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "deepseek-ai. deepseek-ai/DeepSeek-V3.2 model card. Source: https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-tencent-hy3",
    "name": "tencent/Hy3 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/tencent--Hy3.txt",
    "terms_url": "https://huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; root LICENSE begins 'Tencent is pleased to support the open source community by making Tencent Hy3 av'; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "tencent. tencent/Hy3 model card. Source: https://huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-xiaomimimo-mimo-v2-5",
    "name": "XiaomiMiMo/MiMo-V2.5 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/XiaomiMiMo/MiMo-V2.5/blob/63651580ca774f8504f676040460aed3e1244ac1/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/XiaomiMiMo--MiMo-V2.5.txt",
    "terms_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/blob/63651580ca774f8504f676040460aed3e1244ac1/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; no root LICENSE file; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "XiaomiMiMo. XiaomiMiMo/MiMo-V2.5 model card. Source: https://huggingface.co/XiaomiMiMo/MiMo-V2.5/blob/63651580ca774f8504f676040460aed3e1244ac1/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-mistralai-mistral-small-3-2-24b-instruct-2506",
    "name": "mistralai/Mistral-Small-3.2-24B-Instruct-2506 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/mistralai--Mistral-Small-3.2-24B-Instruct-2506.txt",
    "terms_url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "mistralai. mistralai/Mistral-Small-3.2-24B-Instruct-2506 model card. Source: https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the 'Small 3.1 24B Instruct' comparison row (Mistral's own previous checkpoint) of the same tables."
  },
  {
    "id": "card2-mistralai-magistral-small-2509",
    "name": "mistralai/Magistral-Small-2509 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/mistralai--Magistral-Small-2509.txt",
    "terms_url": "https://huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "mistralai. mistralai/Magistral-Small-2509 model card. Source: https://huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card2-google-gemma-4-31b-it",
    "name": "google/gemma-4-31B-it model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/google/gemma-4-31B-it/blob/842da3794eaa0b77d5f08bae87a17459d91ff475/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/google--gemma-4-31B-it.txt",
    "terms_url": "https://huggingface.co/google/gemma-4-31B-it/blob/842da3794eaa0b77d5f08bae87a17459d91ff475/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0; license_link: https://ai.google.dev/gemma/docs/gemma_4_license'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "google. google/gemma-4-31B-it model card. Source: https://huggingface.co/google/gemma-4-31B-it/blob/842da3794eaa0b77d5f08bae87a17459d91ff475/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-qwen-qwen3-coder-480b-a35b-instruct",
    "name": "Qwen/Qwen3-Coder-480B-A35B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-Coder-480B-A35B-Instruct.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0; license_link: https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/main/LICENSE'; root LICENSE begins 'Apache License'; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Qwen. Qwen/Qwen3-Coder-480B-A35B-Instruct model card. Source: https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/9d90cf8fca1bf7b7acca42d3fc9ae694a2194069/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-distil-whisper-distil-large-v3-5",
    "name": "distil-whisper/distil-large-v3.5 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/distil-whisper/distil-large-v3.5/blob/728a7691f3ff1d3d971528d3203a6e9559165d41/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/distil-whisper--distil-large-v3.5.txt",
    "terms_url": "https://huggingface.co/distil-whisper/distil-large-v3.5/blob/728a7691f3ff1d3d971528d3203a6e9559165d41/README.md",
    "quote": "README front-matter declares 'license: mit'. No root LICENSE file at this revision; standard MIT text supplied. Own-lab cells of the Short-Form and Long-Form Evaluation tables in the pinned README, including the Distil-Whisper team's earlier distil-v2 / distil-v3 columns; the OpenAI large-v3 / large-v3-turbo columns are not used (no statement that the card authors ran them), nor are linked blogs or images hosted elsewhere.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Hugging Face (Distil-Whisper). distil-whisper/distil-large-v3.5 model card. Source: https://huggingface.co/distil-whisper/distil-large-v3.5/blob/728a7691f3ff1d3d971528d3203a6e9559165d41/README.md. Licence: MIT. Changes: own-lab numerical cells (distil-large-v3.5 and the earlier distil-v2 / distil-v3 columns) transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-D-2026-10-09.md."
  },
  {
    "id": "first-party-nvidia-asr-cards",
    "name": "NVIDIA ASR model cards (Parakeet / Canary, Hugging Face)",
    "kind": "first-party",
    "license": "CC BY 4.0",
    "match": [
      "huggingface.co/nvidia/parakeet-tdt-0.6b-v3/blob/541d1f99c6b0c3cd0b11a95167540bb8edefd82b/README.md",
      "huggingface.co/nvidia/canary-1b-v2/blob/d455706339a6b32e1aa40f82c713a482a0c938e2/README.md",
      "huggingface.co/nvidia/parakeet-rnnt-1.1b/blob/2acc4c61eede2f52ddefe74935de1930a9064d4a/README.md",
      "huggingface.co/nvidia/parakeet-tdt-0.6b-v2/blob/ae9ad07059c7c739ffaf932226a8fe64ae2620b0/README.md",
      "huggingface.co/nvidia/canary-qwen-2.5b/blob/b1469e1bba1cfe140205529c79c434ca47180960/README.md"
    ],
    "exact": true,
    "scope": "One NVIDIA evaluator for its own ASR cards (Open ASR Leaderboard datasets, greedy decoding, no external LM; multilingual averages after removing punctuation and capitalisation), like first-party-meta-asr-cards / first-party-speechbrain-asr-cards.",
    "license_url": "/licenses/nvidia--parakeet-tdt-0.6b-v2.txt",
    "terms_url": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2/blob/ae9ad07059c7c739ffaf932226a8fe64ae2620b0/README.md",
    "quote": "Each of the five pinned READMEs declares 'license: cc-by-4.0' in its front matter (full notices: /licenses/nvidia--parakeet-tdt-0.6b-v2.txt, /licenses/nvidia--parakeet-tdt-0.6b-v3.txt, /licenses/nvidia--parakeet-rnnt-1.1b.txt, /licenses/nvidia--canary-qwen-2.5b.txt, /licenses/nvidia--canary-1b-v2.txt). Own-lab result cells only; competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "huggingface.co",
    "attribution": "NVIDIA. Model cards for nvidia/parakeet-tdt-0.6b-v2, nvidia/parakeet-tdt-0.6b-v3, nvidia/parakeet-rnnt-1.1b, nvidia/canary-qwen-2.5b, nvidia/canary-1b-v2 on Hugging Face (pinned revisions). Licence: CC BY 4.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-qwen-qwen3-asr-1-7b",
    "name": "Qwen/Qwen3-ASR-1.7B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-ASR-1.7B/blob/7278e1e70fe206f11671096ffdd38061171dd6e5/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-ASR-1.7B.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-ASR-1.7B/blob/7278e1e70fe206f11671096ffdd38061171dd6e5/README.md",
    "quote": "README front-matter declares 'license: apache-2.0'. No root LICENSE file at this revision; standard Apache-2.0 text supplied. Result cells in the pinned README's 'ASR Benchmarks on Public Datasets' table: the Qwen3-ASR-1.7B and -0.6B columns, plus the GPT-4o-Transcribe and Whisper-large-v3 columns, which the Qwen3-ASR Technical Report (arXiv:2601.21337v2, CC BY 4.0, Table 3, identical values) states the authors ran themselves. Gemini-2.5-Pro, Doubao-ASR and the copied FunASR-MLT-Nano column are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Qwen Team, Alibaba. Qwen/Qwen3-ASR-1.7B model card. Source: https://huggingface.co/Qwen/Qwen3-ASR-1.7B/blob/7278e1e70fe206f11671096ffdd38061171dd6e5/README.md. Licence: Apache-2.0. Changes: selected Qwen-run numerical cells (own models and re-run baselines) transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-qwen-qwen3-tts-12hz-1-7b-base",
    "name": "Qwen/Qwen3-TTS-12Hz-1.7B-Base model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base/blob/fd4b254389122332181a7c3db7f27e918eec64e3/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/Qwen--Qwen3-TTS-12Hz-1.7B-Base.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base/blob/fd4b254389122332181a7c3db7f27e918eec64e3/README.md",
    "quote": "README front-matter declares 'license: apache-2.0'. No root LICENSE file at this revision; standard Apache-2.0 text supplied. Selected own-lab result cells in the pinned README only; linked blogs, images hosted elsewhere and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "Qwen Team, Alibaba. Qwen/Qwen3-TTS-12Hz-1.7B-Base model card. Source: https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base/blob/fd4b254389122332181a7c3db7f27e918eec64e3/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09: also the lab's own Qwen3-TTS-12Hz-0.6B-Base row of the same Seed-TTS table; competitor and 25Hz rows not used. Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-E-2026-10-09.md."
  },
  {
    "id": "card2-stepfun-ai-step1x-edit-v1p2",
    "name": "stepfun-ai/Step1X-Edit-v1p2 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/stepfun-ai/Step1X-Edit-v1p2/blob/ca85b97fd19f2235dc0d6fd3633d1319f169e149/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/stepfun-ai--Step1X-Edit-v1p2.txt",
    "terms_url": "https://huggingface.co/stepfun-ai/Step1X-Edit-v1p2/blob/ca85b97fd19f2235dc0d6fd3633d1319f169e149/README.md",
    "quote": "README front-matter declares 'license: apache-2.0'. No root LICENSE file at this revision; standard Apache-2.0 text supplied. Selected own-lab result cells in the pinned README only; linked blogs, images hosted elsewhere and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "StepFun. stepfun-ai/Step1X-Edit-v1p2 model card. Source: https://huggingface.co/stepfun-ai/Step1X-Edit-v1p2/blob/ca85b97fd19f2235dc0d6fd3633d1319f169e149/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-qwen-qwen-image",
    "name": "Qwen/Qwen-Image model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen-Image/blob/75e0b4be04f60ec59a75f475837eced720f823b6/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/Qwen--Qwen-Image.txt",
    "terms_url": "https://huggingface.co/Qwen/Qwen-Image/blob/75e0b4be04f60ec59a75f475837eced720f823b6/LICENSE",
    "quote": "README front-matter declares 'license: apache-2.0'; root LICENSE at the pinned revision is the Apache License 2.0. Selected own-lab result labels from the publisher's benchmark chart embedded in the pinned README (image archived and hashed); linked blogs and competitor bars/columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Qwen Team, Alibaba. Qwen/Qwen-Image model card. Source: https://huggingface.co/Qwen/Qwen-Image/blob/75e0b4be04f60ec59a75f475837eced720f823b6/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-wan-ai-wan2-1-t2v-14b",
    "name": "Wan-AI/Wan2.1-T2V-14B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Wan-AI/Wan2.1-T2V-14B/blob/a064a6c71f5be440641209c07bf2a5ce7a2ff5e4/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/Wan-AI--Wan2.1-T2V-14B.txt",
    "terms_url": "https://huggingface.co/Wan-AI/Wan2.1-T2V-14B/blob/a064a6c71f5be440641209c07bf2a5ce7a2ff5e4/LICENSE.txt",
    "quote": "README front-matter declares 'license: apache-2.0'; root LICENSE.txt at the pinned revision is the Apache License 2.0. Selected own-lab result labels from the publisher's benchmark chart embedded in the pinned README (image archived and hashed); linked blogs and competitor bars/columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "Wan Team, Alibaba. Wan-AI/Wan2.1-T2V-14B model card. Source: https://huggingface.co/Wan-AI/Wan2.1-T2V-14B/blob/a064a6c71f5be440641209c07bf2a5ce7a2ff5e4/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2412.08905v1",
    "name": "Microsoft Research — Phi-4 Technical Report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2412.08905v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2412.08905v1.txt",
    "terms_url": "https://arxiv.org/abs/2412.08905v1",
    "quote": "arXiv abstract page for 2412.08905v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "Marah Abdin et al. (Microsoft Research). Phi-4 Technical Report. arXiv:2412.08905v1. CC BY 4.0. Source: https://arxiv.org/html/2412.08905v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "report-2503.01743v2",
    "name": "Microsoft — Phi-4-Mini Technical Report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2503.01743v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2503.01743v2.txt",
    "terms_url": "https://arxiv.org/abs/2503.01743v2",
    "quote": "arXiv abstract page for 2503.01743v2 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "Microsoft (Abouelenin et al.). Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs. arXiv:2503.01743v2. CC BY 4.0. Source: https://arxiv.org/html/2503.01743v2. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the Phi-3.5-Mini column of Tables 7-8 (same lab) and the unstarred (reproduced) rows of Table 9 including the AIME 2024 column."
  },
  {
    "id": "report-2504.21318v1",
    "name": "Microsoft — Phi-4-reasoning Technical Report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2504.21318v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2504.21318v1.txt",
    "terms_url": "https://arxiv.org/abs/2504.21318v1",
    "quote": "arXiv abstract page for 2504.21318v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "Marah Abdin et al. (Microsoft). Phi-4-reasoning Technical Report. arXiv:2504.21318v1. CC BY 4.0. Source: https://arxiv.org/html/2504.21318v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "report-2503.19786v1",
    "name": "Gemma Team — Gemma 3 Technical Report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2503.19786v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2503.19786v1.txt",
    "terms_url": "https://arxiv.org/abs/2503.19786v1",
    "quote": "arXiv abstract page for 2503.19786v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Gemma Team, Google DeepMind. Gemma 3 Technical Report. arXiv:2503.19786v1. CC BY 4.0. Source: https://arxiv.org/html/2503.19786v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2508.06471v1",
    "name": "GLM-4.5 Team (Zhipu AI & Tsinghua) — GLM-4.5 technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2508.06471v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2508.06471v1.txt",
    "terms_url": "https://arxiv.org/abs/2508.06471v1",
    "quote": "arXiv abstract page for 2508.06471v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "GLM-4.5 Team, Zhipu AI and Tsinghua University. GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models. arXiv:2508.06471v1. CC BY 4.0. Source: https://arxiv.org/html/2508.06471v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2602.15763v2",
    "name": "GLM-5 Team (Zhipu AI) — GLM-5 technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2602.15763v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2602.15763v2.txt",
    "terms_url": "https://arxiv.org/abs/2602.15763v2",
    "quote": "arXiv abstract page for 2602.15763v2 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "GLM-5 Team, Zhipu AI. GLM-5: from Vibe Coding to Agentic Engineering. arXiv:2602.15763v2. CC BY 4.0. Source: https://arxiv.org/html/2602.15763v2. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "report-2504.00698v2",
    "name": "Cohere — Command A technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/pdf/2504.00698v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2504.00698v2.txt",
    "terms_url": "https://arxiv.org/abs/2504.00698v2",
    "quote": "arXiv abstract page for 2504.00698v2 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). No HTML rendering exists; cells transcribed from the pinned PDF via pdftotext -layout. Only the 'Command A' row is used (not Command A Expert/Agentic/Reasoning, Command R7B or external models). The report's statement that model weights are CC-BY-NC does not restrict the report text licence.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "Cohere (Team Cohere et al.). Command A: An Enterprise-Ready Large Language Model. arXiv:2504.00698v2. CC BY 4.0. Source: https://arxiv.org/pdf/2504.00698v2. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Besides the Command A row: internal-reproduction cells of Tables 12-14 (comparators' second value where '/' is printed), the asterisked internal cell of Table 15, and Cohere's own family rows (Command R7B, Command R Refresh, Command R+ Refresh) in Tables 3, 12-14, 21, 22."
  },
  {
    "id": "report-2508.10925v1",
    "name": "OpenAI — gpt-oss-120b & gpt-oss-20b Model Card",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2508.10925v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2508.10925v1.txt",
    "terms_url": "https://arxiv.org/abs/2508.10925v1",
    "quote": "arXiv abstract page for 2508.10925v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "OpenAI. gpt-oss-120b & gpt-oss-20b Model Card. arXiv:2508.10925v1. CC BY 4.0. Source: https://arxiv.org/html/2508.10925v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2605.27605v1",
    "name": "Poolside — Laguna M.1/XS.2 Technical Report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2605.27605v1"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "license_url": "/licenses/report-2605.27605v1.txt",
    "terms_url": "https://arxiv.org/abs/2605.27605v1",
    "quote": "arXiv abstract page for 2605.27605v1 states the article licence: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). Share-alike: adapted extracts must carry the same licence. Only Poolside's own Laguna columns are used; baseline columns (copied from other publishers) are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Poolside Team. Laguna M.1/XS.2 Technical Report. arXiv:2605.27605v1. CC BY-SA 4.0. Source: https://arxiv.org/html/2605.27605v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2512.20848v1",
    "name": "NVIDIA — Nemotron 3 Nano technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2512.20848v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2512.20848v1.txt",
    "terms_url": "https://arxiv.org/abs/2512.20848v1",
    "quote": "arXiv abstract page for 2512.20848v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "NVIDIA. Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning. arXiv:2512.20848v1. CC BY 4.0. Source: https://arxiv.org/html/2512.20848v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2606.15007v1",
    "name": "NVIDIA — Nemotron 3 Ultra technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2606.15007v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2606.15007v1.txt",
    "terms_url": "https://arxiv.org/abs/2606.15007v1",
    "quote": "arXiv abstract page for 2606.15007v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "NVIDIA. Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning. arXiv:2606.15007v1. CC BY 4.0. Source: https://arxiv.org/html/2606.15007v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "report-2603.19220v2",
    "name": "NVIDIA — Nemotron-Cascade 2 technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2603.19220v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2603.19220v2.txt",
    "terms_url": "https://arxiv.org/abs/2603.19220v2",
    "quote": "arXiv abstract page for 2603.19220v2 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Used: the authors' own Nemotron-Cascade-2 cells, and Table 1 baseline cells marked ‡ (baselines NVIDIA evaluated itself with recommended settings). Unmarked baseline cells are official numbers copied from other reports and are excluded. Table 6 LiveCodeBench Pro baseline cells are recorded only with needs_review.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "NVIDIA. Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation. arXiv:2603.19220v2. CC BY 4.0. Source: https://arxiv.org/html/2603.19220v2. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Table 1: own column plus NVIDIA's Nemotron-3-Nano and Nemotron-3-Super columns, and only the '‡'-marked (NVIDIA-evaluated) Qwen3.5-35B-A3B cells."
  },
  {
    "id": "report-2604.12374v1",
    "name": "NVIDIA — Nemotron 3 Super technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2604.12374v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2604.12374v1.txt",
    "terms_url": "https://arxiv.org/abs/2604.12374v1",
    "quote": "arXiv abstract page for 2604.12374v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "NVIDIA. Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning. arXiv:2604.12374v1. CC BY 4.0. Source: https://arxiv.org/html/2604.12374v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2505.00949v5",
    "name": "NVIDIA — Llama-Nemotron: Efficient Reasoning Models",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2505.00949v5"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2505.00949v5.txt",
    "terms_url": "https://arxiv.org/abs/2505.00949v5",
    "quote": "arXiv abstract page for 2505.00949v5 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "NVIDIA. Llama-Nemotron: Efficient Reasoning Models. arXiv:2505.00949v5. CC BY 4.0. Source: https://arxiv.org/html/2505.00949v5. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2507.13264v1",
    "name": "Mistral AI — Voxtral technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2507.13264v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2507.13264v1.txt",
    "terms_url": "https://arxiv.org/abs/2507.13264v1",
    "quote": "arXiv abstract page for 2507.13264v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Own-model result cells are used, plus the Whisper large-v3 and GPT-4o mini Transcribe baseline rows of Tables 3 and 4, which Mistral ran itself ('During evaluation of Voxtral and other models...'; its own E21/E22 10-minute eval sets). Gemini 2.5 Flash, the unversioned 'ElevenLabs Scribe' and the API-only Voxtral Mini Transcribe rows are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Mistral AI. Voxtral. arXiv:2507.13264v1. CC BY 4.0. Source: https://arxiv.org/html/2507.13264v1. Changes: selected own-model and Mistral-run baseline numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2511.22699v5",
    "name": "Tongyi-MAI (Alibaba) — Z-Image technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2511.22699v5"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2511.22699v5.txt",
    "terms_url": "https://arxiv.org/abs/2511.22699v5",
    "quote": "arXiv abstract page for 2511.22699v5 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Z-Image Team, Tongyi Lab, Alibaba. Z-Image: An Efficient Image Generation Foundation Model with Single-Stream Diffusion Transformer. arXiv:2511.22699v5. CC BY 4.0. Source: https://arxiv.org/html/2511.22699v5. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2505.14683v3",
    "name": "ByteDance Seed — BAGEL (Emerging Properties in Unified Multimodal Pretraining)",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2505.14683v3"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2505.14683v3.txt",
    "terms_url": "https://arxiv.org/abs/2505.14683v3",
    "quote": "arXiv abstract page for 2505.14683v3 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Chaorui Deng et al. (ByteDance Seed). Emerging Properties in Unified Multimodal Pretraining. arXiv:2505.14683v3. CC BY 4.0. Source: https://arxiv.org/html/2505.14683v3. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2501.17811v1",
    "name": "DeepSeek-AI — Janus-Pro technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2501.17811v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2501.17811v1.txt",
    "terms_url": "https://arxiv.org/abs/2501.17811v1",
    "quote": "arXiv abstract page for 2501.17811v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Xiaokang Chen et al. (DeepSeek-AI). Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling. arXiv:2501.17811v1. CC BY 4.0. Source: https://arxiv.org/html/2501.17811v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2506.18871v4",
    "name": "BAAI — OmniGen2 technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2506.18871v4"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2506.18871v4.txt",
    "terms_url": "https://arxiv.org/abs/2506.18871v4",
    "quote": "arXiv abstract page for 2506.18871v4 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "Chenyuan Wu et al. (BAAI). OmniGen2: Towards Instruction-Aligned Multimodal Generation. arXiv:2506.18871v4. CC BY 4.0. Source: https://arxiv.org/html/2506.18871v4. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09: Table 4 OmniContext Average for every catalogue model (OmniContext is introduced in this report, so all its cells are the authors' own GPT-4.1-judged runs). Extended 2026-10-09: Scope unchanged in kind (Table 4 OmniContext Average for every catalogue model): adds rows 'OmniGen [98]' (-> omnigen-v1, by lab, BAAI's own earlier model) and 'UNO [97]' (-> uno)."
  },
  {
    "id": "report-2512.07584v1",
    "name": "Meituan LongCat — LongCat-Image technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2512.07584v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2512.07584v1.txt",
    "terms_url": "https://arxiv.org/abs/2512.07584v1",
    "quote": "arXiv abstract page for 2512.07584v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "attribution": "Meituan LongCat Team. LongCat-Image Technical Report. arXiv:2512.07584v1. CC BY 4.0. Source: https://arxiv.org/html/2512.07584v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09: Table 9 CEdit-Bench EN and CN G_O for every catalogue model (CEdit-Bench is introduced in this report, so all its cells are the authors' own GPT-4o-judged runs)."
  },
  {
    "id": "report-2506.03147v4",
    "name": "PKU-YuanGroup — UniWorld-V1 technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2506.03147v4"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2506.03147v4.txt",
    "terms_url": "https://arxiv.org/abs/2506.03147v4",
    "quote": "arXiv abstract page for 2506.03147v4 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Bin Lin et al. (Peking University / Rabbitpre). UniWorld-V1: High-Resolution Semantic Encoders for Unified Visual Understanding and Generation. arXiv:2506.03147v4. CC BY 4.0. Source: https://arxiv.org/html/2506.03147v4. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2409.18869v1",
    "name": "BAAI — Emu3 technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2409.18869v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2409.18869v1.txt",
    "terms_url": "https://arxiv.org/abs/2409.18869v1",
    "quote": "arXiv abstract page for 2409.18869v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Emu3 Team, BAAI. Emu3: Next-Token Prediction is All You Need. arXiv:2409.18869v1. CC BY 4.0. Source: https://arxiv.org/html/2409.18869v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "report-2410.13848v1",
    "name": "DeepSeek-AI — Janus technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2410.13848v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "/licenses/report-2410.13848v1.txt",
    "terms_url": "https://arxiv.org/abs/2410.13848v1",
    "quote": "arXiv abstract page for 2410.13848v1 states the article licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Only the authors' own-model result cells are used; baseline/competitor columns are excluded.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "arxiv.org",
    "attribution": "Chengyue Wu et al. (DeepSeek-AI). Janus: Decoupling Visual Encoding for Unified Multimodal Understanding and Generation. arXiv:2410.13848v1. CC BY 4.0. Source: https://arxiv.org/html/2410.13848v1. Changes: selected own-model numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-thinkingmachines-inkling-small",
    "name": "thinkingmachines/Inkling-Small model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/thinkingmachines/Inkling-Small/blob/8cc5877b44d343f88b92086aa1fb72897950f06a/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/thinkingmachines--Inkling-Small.txt",
    "terms_url": "https://huggingface.co/thinkingmachines/Inkling-Small/blob/8cc5877b44d343f88b92086aa1fb72897950f06a/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0; license_link: https://www.apache.org/licenses/LICENSE-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "thinkingmachines. thinkingmachines/Inkling-Small model card. Source: https://huggingface.co/thinkingmachines/Inkling-Small/blob/8cc5877b44d343f88b92086aa1fb72897950f06a/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-ibm-granite-granite-4-2-30b",
    "name": "ibm-granite/granite-4.2-30b model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ibm-granite/granite-4.2-30b/blob/9e668ce1c538387ef24d3644e9b0606647762636/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/ibm-granite--granite-4.2-30b.txt",
    "terms_url": "https://huggingface.co/ibm-granite/granite-4.2-30b/blob/9e668ce1c538387ef24d3644e9b0606647762636/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "ibm-granite. ibm-granite/granite-4.2-30b model card. Source: https://huggingface.co/ibm-granite/granite-4.2-30b/blob/9e668ce1c538387ef24d3644e9b0606647762636/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-allenai-olmo-3-7b-think",
    "name": "allenai/Olmo-3-7B-Think model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/allenai/Olmo-3-7B-Think/blob/d97e442d7cc678210054dbcc9b440894d62c89a4/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/allenai--Olmo-3-7B-Think.txt",
    "terms_url": "https://huggingface.co/allenai/Olmo-3-7B-Think/blob/d97e442d7cc678210054dbcc9b440894d62c89a4/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "allenai. allenai/Olmo-3-7B-Think model card. Source: https://huggingface.co/allenai/Olmo-3-7B-Think/blob/d97e442d7cc678210054dbcc9b440894d62c89a4/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-A-2026-10-09.md. Detail: Also the 'Olmo 3 Think 7B DPO' column (Ai2's own intermediate checkpoint) on MATH and OMEGA."
  },
  {
    "id": "card2-allenai-olmo-3-1-32b-think",
    "name": "allenai/Olmo-3.1-32B-Think model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/allenai/Olmo-3.1-32B-Think/blob/832c3f543499af8fe68b88359501de9cb7840544/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/allenai--Olmo-3.1-32B-Think.txt",
    "terms_url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think/blob/832c3f543499af8fe68b88359501de9cb7840544/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "allenai. allenai/Olmo-3.1-32B-Think model card. Source: https://huggingface.co/allenai/Olmo-3.1-32B-Think/blob/832c3f543499af8fe68b88359501de9cb7840544/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card2-allenai-olmo-3-7b-instruct",
    "name": "allenai/Olmo-3-7B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/allenai/Olmo-3-7B-Instruct/blob/6e5971d9eba42665f5bd5a0fcf047f299ce1dccc/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/allenai--Olmo-3-7B-Instruct.txt",
    "terms_url": "https://huggingface.co/allenai/Olmo-3-7B-Instruct/blob/6e5971d9eba42665f5bd5a0fcf047f299ce1dccc/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "allenai. allenai/Olmo-3-7B-Instruct model card. Source: https://huggingface.co/allenai/Olmo-3-7B-Instruct/blob/6e5971d9eba42665f5bd5a0fcf047f299ce1dccc/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-B-2026-10-09.md."
  },
  {
    "id": "card2-ibm-granite-granite-4-1-8b",
    "name": "ibm-granite/granite-4.1-8b model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ibm-granite/granite-4.1-8b/blob/1504002f650e656a0a3789d99574df12e3e94ed0/README.md#evaluation-results"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/ibm-granite--granite-4.1-8b.txt",
    "terms_url": "https://huggingface.co/ibm-granite/granite-4.1-8b/blob/1504002f650e656a0a3789d99574df12e3e94ed0/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used. The same README path is already in the ledger as 'first-party-ibm' (a first-party list-price/context fact entry, not a licence review for benchmark cells); this entry is scoped by the URL fragment '#evaluation-results' so benchmark cells resolve to the licence-reviewed entry.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "ibm-granite. ibm-granite/granite-4.1-8b model card. Source: https://huggingface.co/ibm-granite/granite-4.1-8b/blob/1504002f650e656a0a3789d99574df12e3e94ed0/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-ibm-granite-granite-4-0-h-small",
    "name": "ibm-granite/granite-4.0-h-small model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ibm-granite/granite-4.0-h-small/blob/b8c0982bab7fde4eb48110f5a069527c008fab39/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/ibm-granite--granite-4.0-h-small.txt",
    "terms_url": "https://huggingface.co/ibm-granite/granite-4.0-h-small/blob/b8c0982bab7fde4eb48110f5a069527c008fab39/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "ibm-granite. ibm-granite/granite-4.0-h-small model card. Source: https://huggingface.co/ibm-granite/granite-4.0-h-small/blob/b8c0982bab7fde4eb48110f5a069527c008fab39/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-poolside-laguna-m-1",
    "name": "poolside/Laguna-M.1 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/README.md#benchmark-results"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/poolside--Laguna-M.1.txt",
    "terms_url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/LICENSE.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; root LICENSE.md begins 'Apache License'; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used. The same README path is already in the ledger as 'first-party-poolside' (a first-party list-price/context fact entry, not a licence review for benchmark cells); this entry is scoped by the URL fragment '#benchmark-results' so benchmark cells resolve to the licence-reviewed entry.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "poolside. poolside/Laguna-M.1 model card. Source: https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-poolside-laguna-xs-2",
    "name": "poolside/Laguna-XS.2 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/README.md#benchmark-results"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/poolside--Laguna-XS.2.txt",
    "terms_url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/LICENSE.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; root LICENSE.md begins 'Apache License'; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells in the pinned README only; linked blogs, externally hosted images and competitor columns are not used. The same README path is already in the ledger as 'first-party-poolside' (a first-party list-price/context fact entry, not a licence review for benchmark cells); this entry is scoped by the URL fragment '#benchmark-results' so benchmark cells resolve to the licence-reviewed entry.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "poolside. poolside/Laguna-XS.2 model card. Source: https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-huggingfacetb-smollm3-3b",
    "name": "HuggingFaceTB/SmolLM3-3B model card (Evaluation section)",
    "kind": "first-party",
    "match": [
      "huggingface.co/HuggingFaceTB/SmolLM3-3B/blob/a07cc9a04f16550a088caea529712d1d335b0ac1/README.md#evaluation"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/HuggingFaceTB--SmolLM3-3B.txt",
    "terms_url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B/blob/a07cc9a04f16550a088caea529712d1d335b0ac1/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Fragment-scoped to the README '## Evaluation' section (own-model benchmark cells); the same README path without a fragment stays under gap-card-smollm3-3b, whose reviewed scope is context fields.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "attribution": "HuggingFaceTB. HuggingFaceTB/SmolLM3-3B model card. Source: https://huggingface.co/HuggingFaceTB/SmolLM3-3B/blob/a07cc9a04f16550a088caea529712d1d335b0ac1/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "scope": " Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md."
  },
  {
    "id": "card2-zai-org-glm-4-6",
    "name": "zai-org/GLM-4.6 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/GLM-4.6/blob/be72194883d968d7923a07e2f61681ea9a2826d1/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/zai-org--GLM-4.6.txt",
    "terms_url": "https://huggingface.co/zai-org/GLM-4.6/blob/be72194883d968d7923a07e2f61681ea9a2826d1/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; no root LICENSE file; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab printed chart labels only, from the benchmark chart the pinned README embeds (resources/bench_glm46.png, committed by zai-org in its GitHub repository zai-org/GLM-4.5 at 6fb1bbf4e4a5d08f5e8269cd6b87660e56397ff6, root LICENSE Apache-2.0; removed from that repository's main branch on 2025-12-22, archived from the pinned commit). Competitor columns are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "zai-org. zai-org/GLM-4.6 model card. Source: https://huggingface.co/zai-org/GLM-4.6/blob/be72194883d968d7923a07e2f61681ea9a2826d1/README.md. Licence: MIT. Changes: selected own-lab printed chart labels transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-zai-org-glm-5-3-flash",
    "name": "zai-org/GLM-5.3-Flash model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/GLM-5.3-Flash/blob/eb9eb208eb0d988989d07a6a12d0fdeb5f52574a/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/zai-org--GLM-5.3-Flash.txt",
    "terms_url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/eb9eb208eb0d988989d07a6a12d0fdeb5f52574a/LICENSE",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; root LICENSE begins 'MIT License' (Copyright (c) 2026 Z.AI Co., Ltd); grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab printed chart labels only, from the benchmark chart the pinned README embeds (resources/bench_53.png, committed by zai-org in its GitHub repository zai-org/GLM-5 at f6adb53af3c1c620f5a315c127c8875774ff6196, root LICENSE Apache-2.0). Competitor columns and GDPval-AA are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "zai-org. zai-org/GLM-5.3-Flash model card. Source: https://huggingface.co/zai-org/GLM-5.3-Flash/blob/eb9eb208eb0d988989d07a6a12d0fdeb5f52574a/README.md. Licence: MIT. Changes: selected own-lab printed chart labels transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-stepfun-ai-step-3-7-flash",
    "name": "stepfun-ai/Step-3.7-Flash model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/stepfun-ai--Step-3.7-Flash.txt",
    "terms_url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; README License section: 'This project is open-sourced under the Apache 2.0 License'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab printed chart labels from assets/benchmarks.png (committed in the pinned repository and embedded by the README), each confirmed by the same value in the README prose. Unlabelled and competitor bars are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "stepfun-ai. stepfun-ai/Step-3.7-Flash model card. Source: https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/README.md. Licence: Apache-2.0. Changes: selected own-lab printed chart labels transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "card2-xiaomimimo-mimo-v2-5-pro",
    "name": "XiaomiMiMo/MiMo-V2.5-Pro model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "/licenses/XiaomiMiMo--MiMo-V2.5-Pro.txt",
    "terms_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; no root LICENSE file; grant clause: 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...'. Selected own-lab cells of the results table image assets/benchmark.jpg (committed in the pinned repository and embedded by the README). Competitor columns, Elo and rank rows are not used.",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "icon": "huggingface.co",
    "attribution": "XiaomiMiMo. XiaomiMiMo/MiMo-V2.5-Pro model card. Source: https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md. Licence: MIT. Changes: selected own-lab printed table-image cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "speech-stt-afrispeech-multibench-report",
    "name": "AfriSpeech-MultiBench (arXiv:2511.14255v1) · African-accented English ASR benchmark suite",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2511.14255v1",
    "checked": "2026-09-26",
    "attribution": "Gabrial Zencha Ashungafac, Mardhiyah Sanni, Busayo Awobade, Alex Gichamba, Tobi Olatunji (Intron Health); arXiv:2511.14255v1, CC BY 4.0. Table 6 and Table 3 (Parl) WERs transcribed, percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2511.14255v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2511.14255v1 links http://creativecommons.org/licenses/by/4.0/; HTML: 'License: CC BY 4.0'. §3.3 'We evaluate 19 modern ASR systems' zero-shot on the new AfriSpeech-MultiBench subsets (own runs; footnotes give exact model pages). Only AfriSpeech-MultiBench columns are used, not the Open ASR benchmark columns.",
    "license_url": "/licenses/speech-stt-afrispeech-multibench-report.txt"
  },
  {
    "id": "speech-stt-arabic-oual-elm-apache",
    "name": "Open Universal Arabic ASR Leaderboard (Elm Research Center)",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard/blob/667db1e1e40c38f1c5667f7dca1a90eac113231d/README.md",
    "license_url": "/licenses/speech-stt-arabic-oual-elm-apache.txt",
    "checked": "2026-09-26",
    "attribution": "Open Universal Arabic ASR Leaderboard, Elm Research Center (Yingzhi Wang, Anas Alhmoud, Muhammad Alqurishi; arXiv:2412.13788), Hugging Face Space elmresearchcenter/open_universal_arabic_asr_leaderboard, Apache-2.0. Selected Average WER rows from the committed results table in app.py, percent divided by 100; catalogue IDs mapped. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard/blob/667db1e1e40c38f1c5667f7dca1a90eac113231d/app.py"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: apache-2.0'; the leaderboard numbers are committed in the same Space repository as the 'results' dict in app.py (a static table, not loaded from any other repo or API). The Space states the maintainers run the benchmark and open-source the evaluation scripts."
  },
  {
    "id": "speech-stt-bavarian-betthupferl-mit",
    "name": "German dialect ASR Benchmark (Betthupferl; kamaal-08/bavarian-asr-leaderboard)",
    "kind": "leaderboard",
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/kamaal-08/bavarian-asr-leaderboard/blob/1a8d0ac6d092ce3d5fc46d41e9cbc2cd74240bed/README.md",
    "license_url": "/licenses/speech-stt-bavarian-betthupferl-mit.txt",
    "checked": "2026-09-26",
    "attribution": "ASR Benchmark (German dialects, Betthupferl evaluation set), Hugging Face Space kamaal-08/bavarian-asr-leaderboard, MIT. Validated rows from committed data/leaderboard.csv; catalogue IDs mapped. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/spaces/kamaal-08/bavarian-asr-leaderboard/blob/1a8d0ac6d092ce3d5fc46d41e9cbc2cd74240bed/data/leaderboard.csv"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: mit' and states 'Validated results live in data/leaderboard.csv', which 'contains our full-set paper reproductions, additional benchmarks and external fine-tunes' (the Betthupferl authors' own released scores are kept separately in data/paper-results.csv and are not used). The CSV is committed in the Space repository."
  },
  {
    "id": "speech-stt-botnoi-thai-report",
    "name": "BOTNOI Thai telephony ASR report (arXiv:2608.24916v1) · baseline CERs",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2608.24916v1",
    "checked": "2026-09-26",
    "attribution": "Chanameth Boonpramuk, Winn Voravuthikunchai, Songpol Bunyang (Botnoi Group); arXiv:2608.24916v1, CC BY 4.0. Baseline CERs transcribed, percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2608.24916v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2608.24916v1 links http://creativecommons.org/licenses/by/4.0/; HTML: 'License: CC BY 4.0'. The authors 'compare our fine-tuned Canary models against several baselines, including Whisper-large-v3 ... as well as the commercial ElevenLabs Scribe v2 and Google Chirp 3 systems' on their own Thai test sets (own runs; the BOTNOI telephony set is in-house).",
    "license_url": "/licenses/speech-stt-botnoi-thai-report.txt"
  },
  {
    "id": "speech-stt-cl-spanish-asr-apache",
    "name": "Open ASR Leaderboard – Chilean Spanish (idsudd / Alonso Astroza)",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/idsudd/open_asr_leaderboard_cl/blob/7dfb852bee82c9b602548abb33213449c360e31b/README.md",
    "license_url": "/licenses/speech-stt-cl-spanish-asr-apache.txt",
    "checked": "2026-09-26",
    "attribution": "Open ASR Leaderboard – Chilean Spanish, Hugging Face Space idsudd/open_asr_leaderboard_cl (Alonso Astroza), Apache-2.0. Average WER recomputed from committed results.csv exactly as the Space displays it, percent divided by 100; catalogue IDs mapped. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/spaces/idsudd/open_asr_leaderboard_cl/blob/7dfb852bee82c9b602548abb33213449c360e31b/results.csv"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: apache-2.0'; the README says 'The application will load results from results.csv', which is committed in the Space repository with per-dataset WER, sample counts and run job ids (2025-10-26/27) for every evaluated checkpoint."
  },
  {
    "id": "speech-stt-megaasr-report",
    "name": "Mega-ASR (arXiv:2605.19833v1) · adverse-condition ASR baselines (CHiME-4, VOiCES, NOIZEUS)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2605.19833v1",
    "checked": "2026-09-26",
    "attribution": "Zhifei Xie, Kaiyu Pang, Haobin Zhang, Deheng Ye, Xiaobin Hu, Shuicheng Yan, Chunyan Miao; arXiv:2605.19833v1, CC BY 4.0. Table 2 baseline WERs (CHiME-4/VOiCES/NOIZEUS mean) transcribed, percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2605.19833v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2605.19833v1 links http://creativecommons.org/licenses/by/4.0/; HTML: 'License: CC BY 4.0'. §5.1 'We compare against 12 representative systems' on CHiME-4, VOiCES and NOIZEUS; per-condition cells (VOiCES rm1-rm4, NOIZEUS 0-15 dB) are not published by the model makers, so these are the authors' own runs (the paper does not say so in so many words). Table 3 (standard sets, possibly overlapping vendor numbers) is not used.",
    "license_url": "/licenses/speech-stt-megaasr-report.txt"
  },
  {
    "id": "speech-stt-openbench-mit",
    "name": "Argmax OpenBench · open reproducible speech benchmarks (BENCHMARKS.md)",
    "kind": "leaderboard",
    "license": "MIT",
    "terms_url": "https://github.com/argmaxinc/OpenBench/blob/e13108070b9693f2646e4a3e22e4b2d0130d779f/LICENSE",
    "license_url": "/licenses/speech-stt-openbench-mit.txt",
    "checked": "2026-09-26",
    "attribution": "OpenBench, Copyright (c) 2025 argmax, inc., MIT (https://github.com/argmaxinc/OpenBench). Keyword Recognition WER (earnings22-keywords, no-keywords baseline) transcribed by script from the committed BENCHMARKS.md, percent divided by 100; retain the MIT notice. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/argmaxinc/OpenBench/blob/e13108070b9693f2646e4a3e22e4b2d0130d779f/BENCHMARKS.md"
    ],
    "exact": true,
    "icon": "github.com",
    "quote": "Root LICENSE (MIT): 'Copyright (c) 2025 argmax, inc. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute...'. BENCHMARKS.md is a committed documentation file in the repository; no separate data licence or carve-out."
  },
  {
    "id": "speech-stt-persian-triple-threat-apache",
    "name": "Persian ASR Triple Threat Leaderboard (Reza2kn)",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/Reza2kn/PersianASR-TrippleThreat/blob/63e7e3f8b3754a92d6c0ac2a9db84344cf7e49bb/LICENSE",
    "license_url": "/licenses/speech-stt-persian-triple-threat-apache.txt",
    "checked": "2026-09-26",
    "attribution": "Persian ASR Triple Threat Leaderboard, Hugging Face Space Reza2kn/PersianASR-TrippleThreat (Reza Sayar), Apache-2.0. Direct-API and open-weight rows only (OpenRouter-gateway rows excluded); equal-weight mean of the published VisualEars6669 and FLEURS-fa WERs, percent divided by 100; catalogue IDs mapped. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/spaces/Reza2kn/PersianASR-TrippleThreat/blob/63e7e3f8b3754a92d6c0ac2a9db84344cf7e49bb/results.json",
      "huggingface.co/spaces/Reza2kn/PersianASR-TrippleThreat/blob/63e7e3f8b3754a92d6c0ac2a9db84344cf7e49bb/gemini35_transcribe_scores.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Space card README.md declares 'license: apache-2.0' and the Space root holds the Apache License 2.0 LICENSE (commit ba57f25580 'Set Apache-2.0 license metadata'); results.json and gemini35_transcribe_scores.json are committed in the Space and read by the static page. Each row's 'notes' records how it was run: Scribe v2 via the direct ElevenLabs SDK, Gemini 3.5 Transcribe via the official google-genai SDK with model gemini-3.5-transcribe, Qwen3-ASR-0.6B from open weights; OpenRouter-run rows are not used."
  },
  {
    "id": "speech-stt-peste-apache",
    "name": "PESTE · Persian Speech to Text benchmark (fleurs-fa-ir-v1)",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/ArmanJR/PESTE-Benchmark/blob/4ca05cd34f90342d76d12d9e8074e7a26de97e96/LICENSE",
    "license_url": "/licenses/speech-stt-peste-apache.txt",
    "checked": "2026-09-26",
    "attribution": "PESTE - Persian Speech to Text benchmark, Copyright 2026 PESTE contributors (https://github.com/ArmanJR/PESTE-Benchmark), Apache-2.0. Reference texts derive from Google FLEURS fa_ir (CC BY 4.0). Selected rows of the committed generated/leaderboard.csv; CER kept as published fraction. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/ArmanJR/PESTE-Benchmark/blob/4ca05cd34f90342d76d12d9e8074e7a26de97e96/generated/leaderboard.csv"
    ],
    "exact": true,
    "icon": "github.com",
    "quote": "Root LICENSE is the Apache License 2.0; README 'License and citation': 'PESTE code and definitions are Apache-2.0. FLEURS content is CC BY 4.0; see NOTICE.' generated/leaderboard.csv is a committed generated repository file; only aggregate CER/WER values are reused, no FLEURS audio or transcripts."
  },
  {
    "id": "speech-stt-quranic-asr-lb-apache",
    "name": "Quranic ASR Leaderboard (Quran-Lab, held-out recitation set)",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/Quran-Lab/quranic-asr-leaderboard/blob/47d9dd9840ed6897c2c305e4319d16f13360dea0/README.md",
    "license_url": "/licenses/speech-stt-quranic-asr-lb-apache.txt",
    "checked": "2026-09-26",
    "attribution": "Quranic ASR Leaderboard by Quran-Lab (Hugging Face Space Quran-Lab/quranic-asr-leaderboard), Apache-2.0. Overall 600-clip WERs from the committed results.json; percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/spaces/Quran-Lab/quranic-asr-leaderboard/blob/47d9dd9840ed6897c2c305e4319d16f13360dea0/results.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Space card YAML at the pinned revision: 'license: apache-2.0'. results.json (the table app.py renders) and the predictions/*.json run records are committed in the same Space repository with no other licence statement. Quran-Lab ran every system itself (open models re-run 2026-07-02; cloud APIs from its own Quran-Lab/quranic-asr-cloud-rawdata runs), all rescored 2026-08-15 on benchmark v1.1."
  },
  {
    "id": "speech-stt-ru-shmyrev-espeech-mit",
    "name": "Russian ASR Leaderboard (ESpeech; benchmark by Nikolay Shmyrev)",
    "kind": "leaderboard",
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/ESpeech/russian_asr_leaderboard/blob/cc62862c6f7d12840458f868a17a3058da7e639f/README.md",
    "license_url": "/licenses/speech-stt-ru-shmyrev-espeech-mit.txt",
    "checked": "2026-09-26",
    "attribution": "Russian ASR Leaderboard, Hugging Face Space ESpeech/russian_asr_leaderboard; benchmark author Nikolay Shmyrev, Space authors Nikolay Shmyrev and Denis Petrov; MIT. 11-dataset mean WER from the committed static page, percent divided by 100; catalogue IDs mapped. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/spaces/ESpeech/russian_asr_leaderboard/blob/cc62862c6f7d12840458f868a17a3058da7e639f/index.html"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: mit' and names the benchmark author (Nikolay Shmyrev) among the Space authors; all results are embedded in the committed static index.html ('данные встроены' - data embedded), not fetched from elsewhere."
  },
  {
    "id": "speech-stt-sap-dysarthric-report",
    "name": "Dysarthric ASR with clinical context on the Speech Accessibility Project (arXiv:2605.02782v2)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2605.02782v2",
    "checked": "2026-09-26",
    "attribution": "Pehuén Moure, Niclas Pokel, Bilal Bounajma, Yingqiang Gao, Roman Boehringer, Longbiao Cheng, Shih-Chii Liu; arXiv:2605.02782v2, CC BY 4.0. Table 4 zero-shot WER on the SAP rated subset transcribed. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2605.02782v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2605.02782v2 links http://creativecommons.org/licenses/by/4.0/; HTML: 'License: CC BY 4.0'. The authors ran every model themselves (Appendix I, Table 19 lists exact Hugging Face model ids) on their matched SAP rated subset (N=11,218); no copied rows.",
    "license_url": "/licenses/speech-stt-sap-dysarthric-report.txt"
  },
  {
    "id": "speech-stt-turkasr-bench-mit",
    "name": "Turkish ASR Leaderboard (TurkASR-Bench, yagmurtuncer)",
    "kind": "leaderboard",
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/yagmurtuncer/turkish-asr-leaderboard/blob/c27251669fb421135aa8dc2c138b802051775bc8/README.md",
    "license_url": "/licenses/speech-stt-turkasr-bench-mit.txt",
    "checked": "2026-09-26",
    "attribution": "Turkish ASR Leaderboard / TurkASR-Bench, maintainer yagmurtuncer, Hugging Face Space yagmurtuncer/turkish-asr-leaderboard, MIT. Overall corpus WER from committed turkasr_bench.json (version 2026-09-25); catalogue IDs mapped. No endorsement.",
    "verdict": "allowed",
    "match": [
      "huggingface.co/spaces/yagmurtuncer/turkish-asr-leaderboard/blob/c27251669fb421135aa8dc2c138b802051775bc8/turkasr_bench.json"
    ],
    "exact": true,
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: mit'; turkasr_bench.json (all scores, per source and overall) is committed in the same Space repository and read locally by app.py. The README lists the evaluated Hugging Face checkpoints and the protocol (5,200 utterances, corpus WER via jiwer)."
  },
  {
    "id": "speech-stt-voicecodebench-report",
    "name": "VoiceCodeBench (arXiv:2608.28916v2) · structured-token recovery ASR benchmark",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2608.28916v2",
    "checked": "2026-09-26",
    "attribution": "Tyler Baumgartner, Brandon Tai, Lisa Kaelin-Martin, Candice Fan, Luc Debaupte, Bill Wang, Yi Zhong (Besimple AI); arXiv:2608.28916v2, CC BY 4.0. Table 5 batch WER and CTEM transcribed, percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2608.28916v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2608.28916v2 links http://creativecommons.org/licenses/by/4.0/; HTML: 'License: CC BY 4.0'. The authors ran all 19 configurations themselves under a raw-audio-only protocol ('Each prediction artifact records its model identifier and transcript'); the prediction files in the companion HF dataset (MIT for non-audio materials) pin exact API model ids (e.g. model 'gpt-transcribe' via api.openai.com/v1/audio/transcriptions, evaluated 2026-09-01).",
    "license_url": "/licenses/speech-stt-voicecodebench-report.txt"
  },
  {
    "id": "speech-stt-yoruba-cs-report",
    "name": "Switch-aware evaluation of ASR on English-Yoruba code-switched speech (arXiv:2609.11786v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2609.11786v1",
    "checked": "2026-09-26",
    "attribution": "Chibuzor Okocha, Christan Grant; arXiv:2609.11786v1, CC BY 4.0. Table II WER transcribed, percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2609.11786v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2609.11786v1 links http://creativecommons.org/licenses/by/4.0/; HTML: 'License: CC BY 4.0'. §V: 'All systems run zero-shot with greedy decoding ... the same manifests, normalization, and alignment are used for every system' - the authors' own runs on the AfriCodeSwitch validation audio.",
    "license_url": "/licenses/speech-stt-yoruba-cs-report.txt"
  },
  {
    "id": "speech-rime-alpha-bench-ccby",
    "name": "Rime Alpha TTS Benchmark v1.0.0 · SupportBench and ContentBench paired human preference",
    "kind": "dataset",
    "license": "CC BY 4.0",
    "terms_url": "https://github.com/rimelabs/alpha-tts-benchmark/blob/46c080779bf27b23ca2840dab7c2d7f19a945fa7/LICENSING.md",
    "license_url": "/licenses/speech-rime-alpha-bench-ccby.txt",
    "checked": "2026-09-26",
    "attribution": "Rime, AlphaTTSBench v1.0.0 (https://github.com/rimelabs/alpha-tts-benchmark), numerical results licensed CC BY 4.0 (LICENSE-DATA); code MIT. Changes: competitor rows of results/supportbench/supportbench_results.json and results/contentbench/contentbench_results.json selected; sign flipped to competitor-minus-Rime and rescaled (d+2)/4; catalogue IDs mapped from configs/provider_configs.json. Vendor-run comparison against Rime's own Coda model. No endorsement.",
    "verdict": "allowed",
    "match": [
      "github.com/rimelabs/alpha-tts-benchmark/blob/46c080779bf27b23ca2840dab7c2d7f19a945fa7/results/supportbench/supportbench_results.json",
      "github.com/rimelabs/alpha-tts-benchmark/blob/46c080779bf27b23ca2840dab7c2d7f19a945fa7/results/contentbench/contentbench_results.json"
    ],
    "exact": true,
    "icon": "github.com",
    "quote": "LICENSING.md at the pinned commit: 'Original Rime-authored corpus text, benchmark definitions, numerical results, and Rime-owned rights in the dataset compilation are licensed under Creative Commons Attribution 4.0 International (LICENSE-DATA). Attribute Rime and link to this repository and the license; indicate changes.' The two results JSON files are Rime's analyzer outputs over its own frozen listener exports (paired Rime-vs-competitor human judgments collected on Podonos); only the numerical summaries are reused. Listener notes, provider audio and third-party metadata (excluded from the grant) are not used."
  },
  {
    "id": "speech-voicefirst-indic-ccby",
    "name": "Preferences of a Voice-First Nation: pairwise human evaluation of TTS in 10 Indian languages (arXiv:2604.21481v2)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2604.21481v2",
    "license_url": "/licenses/speech-voicefirst-indic-ccby.txt",
    "checked": "2026-09-26",
    "attribution": "Srija Anand, Ashwin Sankar, Mitesh M. Khapra et al.; 'Preferences of a Voice-First Nation: Large-Scale Pairwise Evaluation and Preference Analysis for TTS in Indian Languages'; arXiv:2604.21481v2, CC BY 4.0. Changes: Table 3 overall Bradley-Terry scores transcribed for four catalogue systems. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2604.21481v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The v2 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. The authors generated all audio themselves (default voices, non-streaming) and collected 120K pairwise judgments from 1,915 screened, trained and paid native raters; Table 3 is their own Bradley-Terry fit, not copied rows."
  },
  {
    "id": "speech-instructttseval-ccby",
    "name": "InstructTTSEval: instruction-following TTS benchmark (arXiv:2506.16381v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2506.16381v1",
    "license_url": "/licenses/speech-instructttseval-ccby.txt",
    "checked": "2026-09-26",
    "attribution": "Kexin Huang, Qian Tu, Liwei Fan, Chenchen Yang, Dong Zhang, Shimin Li, Zhaoye Fei, Qinyuan Cheng, Xipeng Qiu (Fudan University); 'InstructTTSEval: Benchmarking Complex Natural-Language Instruction Following in Text-to-Speech Systems'; arXiv:2506.16381v1, CC BY 4.0. Changes: Avg. column of Tables 5 and 6 for catalogue systems, percent divided by 100. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2506.16381v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The v1 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Section 4.2: 'For closed-sourced models, we evaluate gemini-2.5-flash-preview-tts (gemini-flash), gemini-2.5-pro-preview-tts (gemini-pro), gpt-4o-mini-tts, and Hume.' The authors generated and judged all outputs themselves (this is the original source of rows later copied by other papers)."
  },
  {
    "id": "speech-sap-voicerecon-ccby",
    "name": "An Evaluation Framework for TTS Voice Reconstruction (SAP dysarthric speakers; arXiv:2606.21343v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2606.21343v1",
    "license_url": "/licenses/speech-sap-voicerecon-ccby.txt",
    "checked": "2026-09-26",
    "attribution": "Ariadna Sanchez, Christoph Minixhofer, Korin Richmond, Ondřej Klejch, Peter Bell, Simon King (CSTR, University of Edinburgh); 'An Evaluation Framework for Text-to-Speech Voice Reconstruction'; arXiv:2606.21343v1, CC BY 4.0. Changes: Table 2 cells transcribed for four catalogue systems; WER /100; TTSDS /100; BWS log-worth mapped to an Elo-like scale (1000 + 400·log10(e)·log-worth). No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2606.21343v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The v1 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. The authors synthesized 193 utterances per system themselves (17 zero-shot systems, 'selected based on availability of model weights'), ran their own Prolific BWS listening tests and objective metrics (Table 2)."
  },
  {
    "id": "speech-codecmos-accent-ccby",
    "name": "CodecMOS-Accent: listening-test MOS of zero-shot TTS across English accents (arXiv:2603.14328v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2603.14328v1",
    "license_url": "/licenses/speech-codecmos-accent-ccby.txt",
    "checked": "2026-09-26",
    "attribution": "Wen-Chin Huang et al.; 'CodecMOS-Accent: A MOS Benchmark of Resynthesized and TTS Speech from Neural Codecs Across English Accents'; arXiv:2603.14328v1, CC BY 4.0. Changes: Table 1 S-NAT and S-SPK-SIM means for six catalogue voice-cloning systems, normalized (x-1)/4. No endorsement.",
    "verdict": "allowed",
    "match": [
      "arxiv.org/html/2603.14328v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "quote": "The v1 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. The authors generated all 15 voice-cloning systems' samples themselves ('we only chose open-sourced models') and collected 19,600 annotations from 25 crowd listeners (Intergroup); Table 1 is their own system-level result."
  },
  {
    "id": "coverage-download-ministral-3-3b-reasoning",
    "name": "mistralai/Ministral-3-3B-Reasoning-2512-GGUF download metadata",
    "kind": "first-party",
    "license": "Apache-2.0",
    "match": [
      "huggingface.co/mistralai/Ministral-3-3B-Reasoning-2512-GGUF/blob/b432f0a2c144af8a1937b06e2ea7f75ebb4d0b31/Ministral-3-3B-Reasoning-2512-Q4_K_M.gguf"
    ],
    "terms_url": "https://huggingface.co/mistralai/Ministral-3-3B-Reasoning-2512-GGUF/blob/b432f0a2c144af8a1937b06e2ea7f75ebb4d0b31/README.md",
    "license_url": "/licenses/coverage-download-ministral-3-3b-reasoning.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "exact": true,
    "revision": "b432f0a2c144af8a1937b06e2ea7f75ebb4d0b31",
    "quote": "license: apache-2.0",
    "scope": "Publisher-owned downloadable artifacts in the explicitly Apache-2.0 repository. Only file byte lengths from the pinned artifact listing are used, not third-party evaluations or hosted leaderboard data. GB is download size, not total inference memory.",
    "attribution": "mistralai/Ministral-3-3B-Reasoning-2512-GGUF, publisher mistralai. Apache-2.0. Source: https://huggingface.co/mistralai/Ministral-3-3B-Reasoning-2512-GGUF/blob/b432f0a2c144af8a1937b06e2ea7f75ebb4d0b31/Ministral-3-3B-Reasoning-2512-Q4_K_M.gguf. Changes: artifact byte sizes converted to decimal GB, rounded up; command assembled from published usage. No benchmark or runtime-memory claims."
  },
  {
    "id": "coverage-download-ministral-3-8b-reasoning",
    "name": "mistralai/Ministral-3-8B-Reasoning-2512-GGUF download metadata",
    "kind": "first-party",
    "license": "Apache-2.0",
    "match": [
      "huggingface.co/mistralai/Ministral-3-8B-Reasoning-2512-GGUF/blob/4413bc9b7b091ddad28101650d1395f5e045a328/Ministral-3-8B-Reasoning-2512-Q4_K_M.gguf"
    ],
    "terms_url": "https://huggingface.co/mistralai/Ministral-3-8B-Reasoning-2512-GGUF/blob/4413bc9b7b091ddad28101650d1395f5e045a328/README.md",
    "license_url": "/licenses/coverage-download-ministral-3-8b-reasoning.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "exact": true,
    "revision": "4413bc9b7b091ddad28101650d1395f5e045a328",
    "quote": "license: apache-2.0",
    "scope": "Publisher-owned downloadable artifacts in the explicitly Apache-2.0 repository. Only file byte lengths from the pinned artifact listing are used, not third-party evaluations or hosted leaderboard data. GB is download size, not total inference memory.",
    "attribution": "mistralai/Ministral-3-8B-Reasoning-2512-GGUF, publisher mistralai. Apache-2.0. Source: https://huggingface.co/mistralai/Ministral-3-8B-Reasoning-2512-GGUF/blob/4413bc9b7b091ddad28101650d1395f5e045a328/Ministral-3-8B-Reasoning-2512-Q4_K_M.gguf. Changes: artifact byte sizes converted to decimal GB, rounded up; command assembled from published usage. No benchmark or runtime-memory claims."
  },
  {
    "id": "coverage-download-ministral-3-14b-reasoning",
    "name": "mistralai/Ministral-3-14B-Reasoning-2512-GGUF download metadata",
    "kind": "first-party",
    "license": "Apache-2.0",
    "match": [
      "huggingface.co/mistralai/Ministral-3-14B-Reasoning-2512-GGUF/blob/fe3b038f30334729263d860d5dadbaa34e0f2a18/Ministral-3-14B-Reasoning-2512-Q4_K_M.gguf"
    ],
    "terms_url": "https://huggingface.co/mistralai/Ministral-3-14B-Reasoning-2512-GGUF/blob/fe3b038f30334729263d860d5dadbaa34e0f2a18/README.md",
    "license_url": "/licenses/coverage-download-ministral-3-14b-reasoning.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "exact": true,
    "revision": "fe3b038f30334729263d860d5dadbaa34e0f2a18",
    "quote": "license: apache-2.0",
    "scope": "Publisher-owned downloadable artifacts in the explicitly Apache-2.0 repository. Only file byte lengths from the pinned artifact listing are used, not third-party evaluations or hosted leaderboard data. GB is download size, not total inference memory.",
    "attribution": "mistralai/Ministral-3-14B-Reasoning-2512-GGUF, publisher mistralai. Apache-2.0. Source: https://huggingface.co/mistralai/Ministral-3-14B-Reasoning-2512-GGUF/blob/fe3b038f30334729263d860d5dadbaa34e0f2a18/Ministral-3-14B-Reasoning-2512-Q4_K_M.gguf. Changes: artifact byte sizes converted to decimal GB, rounded up; command assembled from published usage. No benchmark or runtime-memory claims."
  },
  {
    "id": "coverage-download-laguna-xs-2",
    "name": "poolside/Laguna-XS.2 download metadata",
    "kind": "first-party",
    "license": "Apache-2.0",
    "match": [
      "huggingface.co/poolside/Laguna-XS.2/tree/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8"
    ],
    "terms_url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/README.md",
    "license_url": "/licenses/coverage-download-laguna-xs-2.txt",
    "verdict": "allowed",
    "checked": "2026-09-26",
    "exact": true,
    "revision": "a397bde04501591d0ac3d74a5d9d0a4ec298ffb8",
    "quote": "license: apache-2.0",
    "scope": "Publisher-owned downloadable artifacts in the explicitly Apache-2.0 repository. Only file byte lengths from the pinned artifact listing are used, not third-party evaluations or hosted leaderboard data. GB is download size, not total inference memory.",
    "attribution": "poolside/Laguna-XS.2, publisher poolside. Apache-2.0. Source: https://huggingface.co/poolside/Laguna-XS.2/tree/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8. Changes: artifact byte sizes converted to decimal GB, rounded up; command assembled from published usage. No benchmark or runtime-memory claims."
  },
  {
    "id": "visual-hpsv2-readme",
    "match": [
      "github.com/tgxs002/HPSv2/blob/866735ecaae999fa714bd9edfa05aa2672669ee3/README.md"
    ],
    "exact": true,
    "name": "tgxs002/HPSv2 repository README, HPS v2 benchmark table",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/tgxs002/HPSv2/blob/866735ecaae999fa714bd9edfa05aa2672669ee3/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-hpsv2-readme.txt",
    "attribution": "Xiaoshi Wu, Yiming Hao, Keqiang Sun, Yixiong Chen, Feng Zhu, Rui Zhao, Hongsheng Li. HPS v2 repository README (tgxs002/HPSv2, commit 866735e), Apache-2.0; paper arXiv:2306.09341. Extracted the 'v2 benchmark' Averaged column for selected models, divided by 100; model labels mapped only to identified checkpoints. No values estimated or rerun. No endorsement.",
    "icon": "github.com",
    "quote": "Repository root LICENSE at commit 866735ecaae999fa714bd9edfa05aa2672669ee3 is the Apache License 2.0, whose Section 1 defines the licensed Work's Source form to include \"documentation source\". The README states the benchmark images were generated by the authors: \"We also provide images generated by models in our benchmark used for evaluation.\" Per-style values are also printed in the CC BY 4.0 paper arXiv:2306.09341v2 Table 7.",
    "scope": "Only the benchmark authors' own 'v2 benchmark' table (images generated by the HPS v2 authors, scored with HPS v2). v2.1 table not used. Withheld rows: SDXL Refiner 0.9, Realistic Vision, Deliberate, ChilloutMix, MajicMix Realistic, Epic Diffusion, DALL·E mini, VQGAN + CLIP, FuseDream, LAFITE (identity/version not established)."
  },
  {
    "id": "visual-tifa",
    "match": [
      "arxiv.org/html/2303.11897v3"
    ],
    "exact": true,
    "name": "TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answering (arXiv v3)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-tifa.txt",
    "attribution": "Yushi Hu, Benlin Liu, Jungo Kasai, Yizhong Wang, Mari Ostendorf, Ranjay Krishna, Noah A. Smith. TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answering, arXiv:2303.11897v3 (17 Aug 2023), CC BY 4.0. Extracted the Appendix Table 4 Overall TIFA score for selected models; 0-100 divided by 100; model labels mapped only to identified checkpoints. No values estimated or rerun. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abs page for 2303.11897v3 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2303.11897v3\".",
    "scope": "Benchmark authors' own TIFA v1.0 evaluation (Sec. 5.2; Appendix B Table 4; Appendix E model details; mPLUG VQA). Withheld: AttnGAN, X-LXMERT and VQ-Diffusion (checkpoints not identified), minDALL-E (min-dalle port running DALL·E Mega weights; not added). The repository leaderboard image is not used."
  },
  {
    "id": "visual-genai-arena",
    "match": [
      "arxiv.org/html/2406.04485v3"
    ],
    "exact": true,
    "name": "GenAI Arena: An Open Evaluation Platform for Generative Models (arXiv v3)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-genai-arena.txt",
    "attribution": "Dongfu Jiang, Max Ku, Tianle Li, Yuansheng Ni, Shizhuo Sun, Rongqi Fan, Wenhu Chen. GenAI Arena: An Open Evaluation Platform for Generative Models, arXiv:2406.04485v3 (6 Aug 2024), CC BY 4.0. Extracted Table 5(a) text-to-image Elo ratings (leaderboard last updated 4 June 2024) for selected models; 95% CIs kept in notes; model labels mapped via Table 2 and the arena's model registry. No values estimated or rerun. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abs page for 2406.04485v3 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2406.04485v3\". v1, v2 and v4 are also CC BY 4.0.",
    "scope": "The authors' own arena ratings from human votes collected on their platform, text-to-image only. Checkpoints from Table 2 and TIGER-AI-Lab/GenAI-Arena arena_elo/generation_model_info.json (commit 52bd731). Withheld: PixArt-Σ (Table 2 says 4K; code uses the 1024-MS checkpoint). The GenAI-Arena website/Space is not used; the v4 (Oct 2024) top-10 snapshot is not imported."
  },
  {
    "id": "visual-imagenhub",
    "match": [
      "arxiv.org/html/2310.01596v4"
    ],
    "exact": true,
    "name": "ImagenHub: Standardizing the evaluation of conditional image generation models (arXiv v4)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-imagenhub.txt",
    "attribution": "Max Ku, Tianle Li, Kai Zhang, Yujie Lu, Xingyu Fu, Wenwen Zhuang, Wenhu Chen. ImagenHub: Standardizing the evaluation of conditional image generation models, arXiv:2310.01596v4 (10 Mar 2024), CC BY 4.0. Extracted the Table 4 text-guided image generation human Overall score for selected models; model labels mapped only to identified checkpoints. No values estimated or rerun. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abs page for 2310.01596v4 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2310.01596v4\".",
    "scope": "The authors' own human evaluation (Sec. 5), text-guided image generation block only. Withheld: Midjourney (version not stated) and 'Stable Diffusion 2.1' (the evaluator's SD wrapper defaults to stable-diffusion-2-base). Editing/subject-driven blocks and automatic metrics not used."
  },
  {
    "id": "visual-evalmuse",
    "match": [
      "arxiv.org/html/2412.18150v2"
    ],
    "exact": true,
    "name": "EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation (arXiv v2)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-evalmuse.txt",
    "attribution": "Shuhao Han, Haotian Fan, Jiachen Fu, Liang Li, Tao Li, Junhui Cui, Yunqiu Wang, Yang Tai, Jingwei Sun, Chunle Guo, Chongyi Li. EvalMuse-40K: A Reliable and Fine-Grained Benchmark with Comprehensive Human Annotations for Text-to-Image Generation Model Evaluation, arXiv:2412.18150v2 (25 Dec 2024), CC BY 4.0. Extracted the Appendix Table 10 overall FGA-BLIP2 score for selected models, stored as (score - 1) / 4; model labels mapped only to identified checkpoints. No values estimated or rerun. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abs page for 2412.18150v2 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2412.18150v2\".",
    "scope": "The authors' own model ranking (200 prompts, their FGA-BLIP2 scorer, Appendix Table 10). Withheld: Dreamina v2.0Pro, HunyuanDiT, Kandinsky3, PixArt-α, PixArt-Σ, PixArt-δ, SDXL-Lightning (version/checkpoint not established). Per-skill columns not used."
  },
  {
    "id": "visual-kolors-readme",
    "match": [
      "github.com/Kwai-Kolors/Kolors/blob/038818d244ed103056abd10f429729a26af4d239/README.md"
    ],
    "exact": true,
    "name": "Kwai-Kolors/Kolors repository README, human assessment",
    "kind": "first-party",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Kwai-Kolors/Kolors/blob/038818d244ed103056abd10f429729a26af4d239/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-kolors-readme.txt",
    "attribution": "Kuaishou Kolors team. Kolors repository README (Kwai-Kolors/Kolors, commit 038818d), Apache-2.0. Extracted the Human Assessment 'Average Overall Satisfaction' values for selected models, stored as (score - 1) / 4; model labels mapped only to identified versions. No values estimated or rerun. No endorsement.",
    "icon": "github.com",
    "quote": "Repository root LICENSE at commit 038818d244ed103056abd10f429729a26af4d239 is the Apache License 2.0 (Section 1 includes documentation source); the README states \"The code of this project is open-sourced under the Apache-2.0 license.\" The separate MODEL_LICENSE covers only the Kolors weights and is not relied on.",
    "scope": "Human Assessment table only (Kuaishou's own evaluation; 'All model results are tested with the April 2024 product versions'); protocol per imgs/Kolors_paper.pdf at the same commit. Kolors row is lab. Withheld: Adobe-Firefly, Stable Diffusion 3, Midjourney-v5 (versions ambiguous). MPS table and other columns not used."
  },
  {
    "id": "visual-conceptmix",
    "match": [
      "arxiv.org/html/2408.14339v1"
    ],
    "exact": true,
    "name": "ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty (arXiv v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-conceptmix.txt",
    "attribution": "Xindi Wu, Dingli Yu, Yangsibo Huang, Olga Russakovsky, Sanjeev Arora. ConceptMix: A Compositional Image Generation Benchmark with Controllable Difficulty, arXiv:2408.14339v1 (26 Aug 2024), CC BY 4.0. Extracted the Table 4 k=2 full-mark score for all eight models (checkpoints from Table 8). No values estimated or rerun. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abs page for 2408.14339v1 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2408.14339v1\".",
    "scope": "Benchmark authors' own runs (Table 4; Table 8 Hugging Face checkpoints; GPT-4o grader; 300 prompts per k). Only the k=2 column is imported; other k levels are correlated tracks of the same experiment."
  },
  {
    "id": "speech-stt-transcribecpp-mit",
    "match": [
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/canary-180m-flash.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/canary-1b-flash.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/canary-1b-v2.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/canary-1b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/canary-qwen-2.5b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/cohere-transcribe-03-2026.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/granite-4.0-1b-speech.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/granite-speech-4.1-2b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/granite-speech-5.0-470m-turboctc.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/moonshine-base.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/moonshine-streaming-medium.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/moonshine-streaming-small.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/moonshine-streaming-tiny.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/moonshine-tiny.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-ctc-0.6b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-ctc-1.1b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-rnnt-0.6b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-rnnt-1.1b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-tdt-0.6b-v2.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-tdt-0.6b-v3.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-tdt-1.1b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/parakeet-tdt_ctc-110m.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/qwen3-asr-0.6b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/qwen3-asr-1.7b.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/voxtral-mini-3b-2507.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/voxtral-mini-4b-realtime-2602.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/voxtral-small-24b-2507.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-base.en.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-base.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-large-v2.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-large-v3-turbo.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-large-v3.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-large.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-medium.en.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-medium.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-small.en.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-small.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-tiny.en.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/whisper-tiny.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/cohere-transcribe-arabic-07-2026.json",
      "github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/catalog/nemotron-3.5-asr-streaming-0.6b.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/parakeet-ultra.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/orukeet.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/parakeet-unified-en-0.6b.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/nemotron-speech-streaming-en-0.6b.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/multitalker-parakeet-streaming-0.6b-v1.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/parakeet-tdt_ctc-1.1b.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/granite-speech-4.1-2b-nar.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/granite-speech-4.1-2b-plus.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/granite-speech-5.0-470m-turboctc-nc.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/moss-transcribe-diarize.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/fun-asr-nano-2512.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/fun-asr-mlt-nano-2512.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/sensevoice-small.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/breeze-asr-25.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/medasr.json",
      "github.com/handy-computer/transcribe.cpp/blob/18947cf405f957195dd457db4e34c1cff9e88889/catalog/parakeet-primeline.json"
    ],
    "exact": true,
    "name": "transcribe.cpp model catalog · LibriSpeech test-clean WER of validated full-precision ports",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/handy-computer/transcribe.cpp/blob/c1fee503219ea82dc2e42917150198a31b2ecc2a/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "license_url": "/licenses/speech-stt-transcribecpp-mit.txt",
    "attribution": "transcribe.cpp, Copyright (c) 2026 The transcribe.cpp authors, MIT (https://github.com/handy-computer/transcribe.cpp). LibriSpeech test-clean corpus WER of the F32 / native-BF16 / F16 tier transcribed by script from the committed catalog/*.json at c1fee50, percent divided by 100; retain the MIT notice. No endorsement.",
    "icon": "github.com",
    "quote": "Root LICENSE at the pinned commit (MIT): 'Copyright (c) 2026 The transcribe.cpp authors. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies...'. README: 'transcribe.cpp is MIT-licensed' and 'catalog/ is the source of truth for model metadata, accuracy, and performance'. The catalog JSON files are committed repository files; THIRD-PARTY-LICENSES.md lists only ggml and miniz (both MIT); no separate data licence or carve-out."
  },
  {
    "id": "speech-stt-srb-report",
    "match": [
      "arxiv.org/html/2403.07937v3"
    ],
    "exact": true,
    "name": "Speech Robust Bench (arXiv:2403.07937v3) · clean-speech WER of 25+ open ASR models",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2403.07937v3",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-stt-srb-report.txt",
    "attribution": "Muhammad A. Shah, David Solans, Mikko A. Heikkilä, Bhiksha Raj, Nicolas Kourtellis, 'Speech Robust Bench: A Robustness Benchmark For Speech Recognition', arXiv:2403.07937v3 (9 Dec 2024), CC BY 4.0. Table 7 clean-speech WER (LibriSpeech, TEDLIUM) transcribed by script, percent divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2403.07937v3 links http://creativecommons.org/licenses/by/4.0/ and the HTML states 'License: CC BY 4.0'. §4.1: 'For English, we evaluate Whisper ... Wav2Vec-2.0 ... HuBERT ... Nvidia Canary ... Nvidia Parakeet RNN-T and CTC ... MMS ... We used the Huggingface implementations where available'; Table 5 names each checkpoint. All rows are the authors' own runs."
  },
  {
    "id": "speech-stt-picovoice-apache",
    "match": [
      "github.com/Picovoice/speech-to-text-benchmark/blob/43e7689f013694e9ba9910695643af9c579ca8d2/README.md",
      "github.com/Picovoice/speech-to-text-benchmark/blob/aea7914706d44d3163c1c0283e05700900abe7af/README.md"
    ],
    "exact": true,
    "name": "Picovoice speech-to-text-benchmark · English batch-engine WER (README)",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Picovoice/speech-to-text-benchmark/blob/43e7689f013694e9ba9910695643af9c579ca8d2/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "license_url": "/licenses/speech-stt-picovoice-apache.txt",
    "attribution": "Picovoice speech-to-text-benchmark (https://github.com/Picovoice/speech-to-text-benchmark), Apache License 2.0. English batch-engine 'Average' WER for the Whisper rows transcribed by script from README.md at 43e7689, percent divided by 100; retain the Apache-2.0 notice. No endorsement.",
    "icon": "github.com",
    "quote": "Root LICENSE at the pinned commit is the Apache License, Version 2.0: 'Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works'. README.md (with the results tables) is a committed file of the Work; no separate data licence or carve-out."
  },
  {
    "id": "video-genai-arena-space",
    "match": [
      "huggingface.co/spaces/TIGER-Lab/GenAI-Arena/blob/4197b3a6b50aca8fae6c01d5968f08c3537024d1/arena_elo/results/20250322/video_generation_leaderboard.csv"
    ],
    "exact": true,
    "name": "GenAI-Arena (TIGER-Lab) text-to-video leaderboard",
    "kind": "leaderboard",
    "license": "MIT",
    "terms_url": "https://huggingface.co/spaces/TIGER-Lab/GenAI-Arena/blob/4197b3a6b50aca8fae6c01d5968f08c3537024d1/README.md",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/video-genai-arena-space.txt",
    "attribution": "GenAI-Arena by TIGER-Lab (Dongfu Jiang, Max Ku, Tianle Li, Yuansheng Ni, Shizhuo Sun, Rongqi Fan, Wenhu Chen; arXiv:2406.04485), Hugging Face Space TIGER-Lab/GenAI-Arena at revision 4197b3a, MIT (Space card). Selected rows of the committed text-to-video leaderboard CSV (anonymous-vote Elo, rounded to 0.1); catalogue IDs mapped via the Space's model registry and VideoGenHub. No endorsement.",
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: mit'; arena_elo/results/20250322/video_generation_leaderboard.csv (and the identical results/latest copy) are committed in the same Space repository with no other licence statement. The Space's own arena_elo scripts compute the Elo from votes collected by the platform; the paper (arXiv 2406.04485, CC BY 4.0) describes the same leaderboard.",
    "scope": "Text-to-video leaderboard CSV only (snapshot 2025-03-22). Rows whose model identity cannot be tied to one checkpoint (fal SVD/AnimateDiff pipelines, Open-Sora 1.1 stage-2) are not imported."
  },
  {
    "id": "video-love-aigve60k",
    "match": [
      "arxiv.org/html/2505.12098v1"
    ],
    "exact": true,
    "name": "LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/video-love-aigve60k.txt",
    "attribution": "Jiarui Wang, Huiyu Duan, Ziheng Jia, Yu Zhao, Woo Yi Yang, Zicheng Zhang, Zijian Chen, Juntong Wang et al. (Shanghai Jiao Tong University). LOVE / AIGVE-60K, arXiv:2505.12098v1, CC BY 4.0. Extracted the Overall column of Tables 8 and 9 (human perception and correspondence MOS) for models with an identifiable release; values divided by 100. No values estimated or rerun. No endorsement; provided as-is.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2505.12098v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0); the HTML of v1 states 'License: CC BY 4.0'.",
    "scope": "Authors' own generation of 58,500 videos from 30 T2V models and own subjective study. Only human MOS tables (8, 9); the LOVE metric's automatic scores and task accuracy (Table 10) are not imported. Unversioned commercial products and ambiguous open checkpoints omitted."
  },
  {
    "id": "video-worldmodelbench",
    "match": [
      "arxiv.org/html/2502.20694v1"
    ],
    "exact": true,
    "name": "WorldModelBench: Judging Video Generation Models As World Models",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/video-worldmodelbench.txt",
    "attribution": "Dacheng Li, Yunhao Fang, Yukang Chen, Shuo Yang, Shiyi Cao, Justin Wong, Michael Luo, Xiaolong Wang, Hongxu Yin, Joseph E. Gonzalez, Ion Stoica, Song Han, Yao Lu. WorldModelBench, arXiv:2502.20694v1, CC BY 4.0. Extracted Table 3 human-annotation Total scores for models with an identifiable checkpoint/release; divided by 10. No values estimated or rerun. No endorsement; provided as-is.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2502.20694v1 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0); the HTML of v1 states 'License: CC BY 4.0'.",
    "scope": "Authors' own generations (settings in Appendix 8.3) and own human annotation (67K labels). Table 3 Total only; judge-model scores (Table 9), Elo figure and Hard subset not imported. Kling (mode), MiniMax (version), Runway ('runway-3.0'), Mochi-official and the Open-Sora 1.2 I2V duplicate omitted."
  },
  {
    "id": "video-seaweed-7b",
    "match": [
      "arxiv.org/html/2504.08685v2"
    ],
    "exact": true,
    "name": "Seaweed-7B: Cost-Effective Training of Video Generation Foundation Model",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/video-seaweed-7b.txt",
    "attribution": "Team Seawead (ByteDance Seed): Ceyuan Yang, Zhijie Lin, Yang Zhao, Shanchuan Lin, Zhibei Ma, Haoyuan Guo, Hao Chen et al. Seaweed-7B, arXiv:2504.08685v2, CC BY 4.0. Extracted Table 3 image-to-video Elo (MagicArena human evaluation); model labels mapped only where the release/mode is identifiable. No values estimated or rerun. No endorsement; provided as-is.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2504.08685v2 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version (v1 and v2 both CC BY 4.0).",
    "scope": "ByteDance's own MagicArena human pairwise evaluation reported in the report (Table 3, I2V). Seaweed-7B row is lab; Wan 2.1 and HunyuanVideo rows omitted (hosted API vs open checkpoint unclear). T2V Elo appears only in a figure and is not imported."
  },
  {
    "id": "video-videoverse",
    "match": [
      "arxiv.org/html/2510.08398v4"
    ],
    "exact": true,
    "name": "VideoVerse: Does Your T2V Generator Have World Model Capability to Synthesize Videos?",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/video-videoverse.txt",
    "attribution": "Zeqing Wang, Xinyu Wei, Bairui Li, Zhen Guo, Jinrui Zhang, Hongyang Wei, Keze Wang, Lei Zhang (Sun Yat-sen University, Hong Kong Polytechnic University, Tsinghua University). VideoVerse, arXiv:2510.08398v4, CC BY 4.0. Extracted Table 1 Overall points divided by the 1,608 evaluation items; model labels mapped per Appendix D. No values estimated or rerun. No endorsement; provided as-is.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2510.08398v4 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) (v1-v4 all CC BY 4.0); the HTML of v4 states 'License: CC BY 4.0'.",
    "scope": "Authors' own generations (open models on A800; Veo-3-fast and Sora-2 standard via official APIs) and own Gemini 2.5 Pro evaluation. Table 1 Overall only. SkyReels-V2 (checkpoint), Wan2.1-14B (linked checkpoint is I2V), MiniMax-Hailuo (version) and the 10 s CogVideoX1.5 setting omitted."
  },
  {
    "id": "video-chronomagic-bench",
    "match": [
      "arxiv.org/html/2406.18522v2"
    ],
    "exact": true,
    "name": "ChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/video-chronomagic-bench.txt",
    "attribution": "Shenghai Yuan, Jinfa Huang, Yongqi Xu, Yaoyang Liu, Shaofeng Zhang, Yujun Shi, Ruijie Zhu, Xinhua Cheng, Jiebo Luo, Li Yuan (Peking University, University of Rochester). ChronoMagic-Bench, arXiv:2406.18522v2, CC BY 4.0. Extracted the GPT4o-MTScore column of Table 4 (ChronoMagic-Bench-150) for models with an identifiable release; 1-5 values divided by 5. No values estimated or rerun. No endorsement; provided as-is.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2406.18522v2 links the licence http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0) for this version (v1 and v2 both CC BY 4.0).",
    "scope": "Authors' own generations and own GPT-4o scoring on the 150-prompt subset (Table 4). The self-submission leaderboard (GitHub CSV / HF Space) is not used. Zero-shot/distilled pipelines, the authors' own models (MagicTime, Open-Sora Plan) and rows whose checkpoint is not identifiable (AnimateDiff V3 base, Latte vs Latte-1, EasyAnimate V3 '720P') omitted."
  },
  {
    "id": "eleutherai-gpt-j-6b-card",
    "name": "EleutherAI GPT-J-6B model card — EleutherAI lm-evaluation-harness runs",
    "kind": "first-party",
    "match": [
      "huggingface.co/EleutherAI/gpt-j-6b/blob/47e169305d2e8376be1d31e765533382721b2cc1/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/EleutherAI/gpt-j-6b/blob/47e169305d2e8376be1d31e765533382721b2cc1/README.md",
    "license_url": "/licenses/eleutherai-gpt-j-6b-card-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "EleutherAI, GPT-J-6B model card (huggingface.co/EleutherAI/gpt-j-6b), Apache-2.0. Selected EleutherAI-run evaluation cells; percentages divided by 100. No endorsement.",
    "quote": "Model-card YAML at the pinned revision: `license: apache-2.0`. Table footnote: '* Evaluation numbers reported by their respective authors. All other numbers are provided by running lm-evaluation-harness either with released weights or with API access.' Only non-* rows are used.",
    "scope": "Non-* rows of the 'Evaluation results' table (8 models x 4 zero-shot tasks).",
    "icon": "huggingface.co"
  },
  {
    "id": "eleutherai-gpt-neo-2-7b-card",
    "name": "EleutherAI GPT-Neo 2.7B model card — EleutherAI evaluation-harness runs",
    "kind": "first-party",
    "match": [
      "huggingface.co/EleutherAI/gpt-neo-2.7B/blob/e24fa291132763e59f4a5422741b424fb5d59056/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://huggingface.co/EleutherAI/gpt-neo-2.7B/blob/e24fa291132763e59f4a5422741b424fb5d59056/README.md",
    "license_url": "/licenses/eleutherai-gpt-neo-2-7b-card-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "EleutherAI, GPT-Neo 2.7B model card (huggingface.co/EleutherAI/gpt-neo-2.7B), MIT. Selected evaluation cells; percentages divided by 100. No endorsement.",
    "quote": "Model-card YAML at the pinned revision: `license: mit`. Card: 'All evaluations were done using our evaluation harness.'",
    "scope": "MathQA and PubMedQA cells (4 models).",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2204.06745v1",
    "name": "GPT-NeoX-20B: An Open-Source Autoregressive Language Model — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2204.06745v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2204.06745v1",
    "license_url": "/licenses/paper-2204.06745v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Sid Black, Stella Biderman, Eric Hallahan et al. (2022). GPT-NeoX-20B: An Open-Source Autoregressive Language Model. arXiv:2204.06745v1. https://arxiv.org/abs/2204.06745v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2204.06745v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Appendix D Table 2 (zero-shot natural-language tasks for GPT-J-6B, GPT-NeoX-20B and the GPT-3 API models Ada/Babbage/Curie/DaVinci), tasks LAMBADA, HellaSwag, WinoGrande, SciQ, PIQA, TriviaQA, ARC Easy/Challenge, OpenBookQA. The authors ran every column with the EleutherAI harness (GPT-3 via the OpenAI API). FairSeq tables, five-shot tables, ANLI/WSC/LogiQA/PROST/QA4MRE/HeadQA rows and Hendrycks/arithmetic tables not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2304.01373v2",
    "name": "Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2304.01373v2"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2304.01373v2",
    "license_url": "/licenses/paper-2304.01373v2-CC-BY-SA-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Stella Biderman, Hailey Schoelkopf, Quentin Anthony et al. (2023). Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling. arXiv:2304.01373v2. https://arxiv.org/abs/2304.01373v2\nCC BY-SA 4.0: https://creativecommons.org/licenses/by-sa/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2304.01373v2 links the licence 'http://creativecommons.org/licenses/by-sa/4.0/' in its licence field, and the HTML version header states 'License: CC BY-SA 4.0'. Share-alike: rows adapted from this paper must be offered under CC BY-SA 4.0 (see adapted_data_file).",
    "scope": "Appendix G zero-shot tables: Table 7 (fully-trained standard Pythia suite), Table 15 (BLOOM 560M-3B) and Table 16 (OPT 125M-66B); rows LAMBADA (OpenAI), PIQA, WinoGrande, ARC-Easy, ARC-Challenge, SciQ. The paper states the authors 'run all evaluations ourselves instead of copying claimed results from previous papers'. Deduplicated-suite, 93k-step and five-shot tables and the WSC/LogiQA rows not imported.",
    "icon": "arxiv.org",
    "adapter_license": "CC-BY-SA-4.0",
    "publication_requirement": "Share-alike: retain attribution, change notice and licence when displaying or serving these rows; adapted rows must be offered under CC BY-SA 4.0.",
    "adapter_license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "adapted_data_file": "/data/llm-new-bysa.json"
  },
  {
    "id": "paper-2402.00838v4",
    "name": "OLMo: Accelerating the Science of Language Models — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2402.00838v4"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2402.00838v4",
    "license_url": "/licenses/paper-2402.00838v4-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Dirk Groeneveld, Iz Beltagy, Pete Walsh et al. (2024). OLMo: Accelerating the Science of Language Models. arXiv:2402.00838v4. https://arxiv.org/abs/2402.00838v4\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2402.00838v4 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 3 (zero-shot, 8 core tasks: ARC Challenge/Easy, BoolQ, HellaSwag, OpenBookQA, PIQA, SciQ, WinoGrande) for OLMo-1B, OLMo-7B and the baselines Pythia 1B, Falcon-7B, LLaMA 7B, Llama 2 7B, MPT-7B, Pythia 6.9B, RPJ-INCITE-7B; the paper states it uses 'our fixed evaluation pipeline to compare results against publicly available models' (Catwalk, rank classification). StableLM 1.6B and TinyLlama 1.1B rows, the avg. column and Table 7 not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2309.05463v1",
    "name": "Textbooks Are All You Need II: phi-1.5 technical report — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2309.05463v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2309.05463v1",
    "license_url": "/licenses/paper-2309.05463v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Yuanzhi Li, Sébastien Bubeck, Ronen Eldan, Allie Del Giorno, Suriya Gunasekar, Yin Tat Lee (2023). Textbooks Are All You Need II: phi-1.5 technical report. arXiv:2309.05463v1. https://arxiv.org/abs/2309.05463v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2309.05463v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2 (WinoGrande, ARC-Easy, ARC-Challenge, BoolQ, SIQA) and Table 3 (PIQA, HellaSwag, OpenbookQA) zero-shot accuracies, LM-Eval Harness, for phi-1.5 and the re-run baselines Llama2-7B, Llama-7B, MPT-7B, Falcon-7B, Falcon-rw-1.3B, OPT-1.3B, GPT-Neo-2.7B, GPT2-XL-1.5B (Figure 1 caption: 'the numbers are from our own evaluation pipeline, to ensure consistency between models'). Vicuna-13B and unreleased phi-1.5-web rows, the MMLU and SQUAD columns of Table 3 and Table 4 (math/code; contains copied rows) not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2302.13971v1",
    "name": "LLaMA: Open and Efficient Foundation Language Models — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2302.13971v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2302.13971v1",
    "license_url": "/licenses/paper-2302.13971v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Hugo Touvron, Thibaut Lavril, Gautier Izacard et al. (2023). LLaMA: Open and Efficient Foundation Language Models. arXiv:2302.13971v1. https://arxiv.org/abs/2302.13971v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2302.13971v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "LLaMA 7B/13B/33B/65B rows only (Meta's own measurements of its own models): Table 3 zero-shot common-sense tasks, Table 7 MATH and GSM8k (single-sample columns, not maj1@k), Table 8 HumanEval and MBPP pass@1, Table 9 MMLU five-shot average. Rows for GPT-3, Gopher, Chinchilla, PaLM, Minerva, LaMDA and GPT-NeoX in these tables are copied from other papers and are not imported; NaturalQuestions, TriviaQA, RACE and bias/toxicity tables not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2211.05100v4",
    "name": "BLOOM: A 176B-Parameter Open-Access Multilingual Language Model — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2211.05100v4"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2211.05100v4",
    "license_url": "/licenses/paper-2211.05100v4-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "BigScience Workshop (2023). BLOOM: A 176B-Parameter Open-Access Multilingual Language Model. arXiv:2211.05100v4. https://arxiv.org/abs/2211.05100v4\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2211.05100v4 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 9 (HumanEval) pass@1 column for the pretrained BLOOM models 560M-176B only (BigScience's own measurements). The caption states non-BLOOM rows (GPT-Neo, GPT-J, GPT-NeoX, Codex) come from prior work, so they are not imported; BLOOMZ rows and pass@10/100 not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2210.11416v5",
    "name": "Scaling Instruction-Finetuned Language Models — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2210.11416v5"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2210.11416v5",
    "license_url": "/licenses/paper-2210.11416v5-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Hyung Won Chung, Le Hou, Shayne Longpre et al. (2022). Scaling Instruction-Finetuned Language Models. arXiv:2210.11416v5. https://arxiv.org/abs/2210.11416v5\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2210.11416v5 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 5 Flan-T5 Small/Base/Large/XL/XXL rows, MMLU 'Direct' (five-shot) and BBH 'Direct' (three-shot) columns (Google's own evaluation of its released Flan-T5 checkpoints). T5 rows (LM-adapted T5, not the original T5 checkpoints), PaLM/Flan-PaLM/U-PaLM rows, CoT, TyDiQA and MGSM columns not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2306.11644v2",
    "name": "Textbooks Are All You Need — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2306.11644v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2306.11644v2",
    "license_url": "/licenses/paper-2306.11644v2-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Suriya Gunasekar, Yi Zhang, Jyoti Aneja et al. (2023). Textbooks Are All You Need. arXiv:2306.11644v2. https://arxiv.org/abs/2306.11644v2\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2306.11644v2 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 1 phi-1 row only (Microsoft's own HumanEval and MBPP pass@1 for its model). All other rows are self-reported scores of other models copied by the authors and are not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2312.06281v2",
    "name": "EQ-Bench: An Emotional Intelligence Benchmark for Large Language Models — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2312.06281v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2312.06281v2",
    "license_url": "/licenses/paper-2312.06281v2-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Samuel J. Paech (2024). EQ-Bench: An Emotional Intelligence Benchmark for Large Language Models. arXiv:2312.06281v2. https://arxiv.org/abs/2312.06281v2\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2312.06281v2 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 1 (EQ-Bench v1 scores, author's own runs; Section 4.1 headline table), used consistently for every row. Rows: OpenAI text-davinci-001 (old-llm lane) and, in the chat-code lane, gpt-4-0613, gpt-4-0314, gpt-3.5-turbo-0613 ('gpt-3.5-0613'), gpt-3.5-turbo-0301, text-davinci-003, text-davinci-002, Llama-2-70b/13b/7b-chat, Yi-34B-Chat, Qwen-14B-Chat, openchat_3.5, vicuna-33b-v1.3, Koala 13B. Table 1 is preferred over the appendix correlation Tables 3-5, which print different values for several models (e.g. gpt-4-0314 60.73 vs 53.39) and swap text-davinci-002/003; Table 6 repeat runs agree with Table 1. Not imported: 'Anthropic Claude2' (snapshot not named), Curie/ADA (simplified prompt, historical set), Babbage (fail), community fine-tunes and Qwen-72B-Chat (not in the catalogue).",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2310.06825v1",
    "name": "Mistral 7B — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2310.06825v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2310.06825v1",
    "license_url": "/licenses/paper-2310.06825v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "attribution": "Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch et al. (2023). Mistral 7B. arXiv:2310.06825v1. https://arxiv.org/abs/2310.06825v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2310.06825v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2: authors' own evaluations of LLaMA 2 7B/13B, Mistral 7B and the code-pretrained Code Llama 7B base checkpoint (not Instruct/Python). Columns MMLU, HellaSwag, WinoGrande, PIQA, ARC-e/c, Natural Questions, TriviaQA, HumanEval, MBPP, MATH and GSM8K; benchmark settings remain as reported. Chat-model Table 3 and figures excluded. Code Llama and Natural Questions scope reviewed 2026-10-06. Extended 2026-10-09 (pass 7): also comparator columns that the document shows the authors measured themselves (same-lab sibling columns as by: lab; other makers' models as by: independent); copied, uncredited or third-party-credited cells are still excluded. Evidence per source: docs/research/pass7-lone-C-2026-10-09.md.",
    "icon": "arxiv.org"
  },
  {
    "id": "acl-2021-emnlp-main-564",
    "name": "Surface Form Competition (Holtzman et al., EMNLP 2021) — authors' own GPT-2 runs",
    "kind": "paper",
    "match": [
      "aclanthology.org/2021.emnlp-main.564.pdf"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://aclanthology.org/2021.emnlp-main.564/",
    "license_url": "/licenses/acl-2021-emnlp-main-564-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Ari Holtzman, Peter West, Vered Shwartz, Yejin Choi and Luke Zettlemoyer; ©2021 Association for Computational Linguistics. Surface Form Competition: Why the Highest Probability Answer Isn’t Always Right. Proceedings of EMNLP 2021, pages 7038–7051.\nSource: https://aclanthology.org/2021.emnlp-main.564.pdf\nLicence: CC BY 4.0 — https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "ACL Anthology landing page footer for 2021.emnlp-main.564: 'Materials published in or after 2016 are licensed on a Creative Commons Attribution 4.0 International License.' The paper (EMNLP 2021) is an ACL publication; Section 4.1: 'We use GPT-2 via the HuggingFace Transformers library ... We do not finetune any models'.",
    "scope": "Appendix A, Table 6 ('Multiple Choice Accuracy on GPT-2'), 'LM' scoring column (standard unnormalised conditional log-probability) for the four released GPT-2 models 125M/350M/760M/1.6B, rows COPA, StoryCloze, HellaSwag, ARC-Easy, ARC-Challenge, OpenBookQA, CommonsenseQA. The authors ran GPT-2 themselves via Hugging Face Transformers (Section 4.1). Unc/Avg/PMI_DC/CC columns, RACE-M/H, BoolQ, RTE, CB, SST-2/5, AG News, TREC rows, and Table 5 (GPT-3 API models labelled only by assumed parameter counts 2.7B/6.7B/13B/175B, so their exact API identities are not established) not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "speech-f5tts-report-ccby",
    "match": [
      "arxiv.org/html/2410.06885v1"
    ],
    "exact": true,
    "name": "F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching (arXiv:2410.06885v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2410.06885v1",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-f5tts-report-ccby.txt",
    "attribution": "Yushen Chen et al. (SJTU / X-LANCE); 'F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching'; arXiv:2410.06885v1, CC BY 4.0. Changes: Tables 1-2: WER and SIM-o for the authors' F5-TTS and reproduced E2 TTS and their re-runs of CosyVoice-300M and FireRedTTS (rows marked * as copied are excluded); percentages divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v1 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Sec. 5: 'We report the average score of three random seed generation results with our model and open-sourced baselines'; Appendix A: FireRedTTS and CosyVoice 'We use the official code and pre-trained checkpoint' (github.com/FireRedTeam/FireRedTTS, huggingface.co/model-scope/CosyVoice-300M); Table 2 caption: '* denotes scores reported in baseline papers' (those rows excluded)."
  },
  {
    "id": "speech-distar-report-ccby",
    "match": [
      "arxiv.org/html/2510.12210v2"
    ],
    "exact": true,
    "name": "DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation (arXiv:2510.12210v2)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2510.12210v2",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-distar-report-ccby.txt",
    "attribution": "DiSTAR authors (arXiv:2510.12210v2); 'DiSTAR: Diffusion over a Scalable Token Autoregressive Representation for Speech Generation'; arXiv:2510.12210v2, CC BY 4.0. Changes: Table 1 WER/SIM/UTMOS and Table 2 SMOS for re-run open baselines only (IndexTTS, E2 TTS, F5-TTS v1, CosyVoice 2, FireRedTTS); DiTAR rows copied from its paper and the authors' unreleased DiSTAR models excluded; WER divided by 100, MOS normalised (x-1)/4. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v2 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Appendix E lists each open baseline with 'We use the official code and pre-trained checkpoint' (CosyVoice2-0.5B, IndexTeam/Index-TTS, SWivid/F5-TTS, FireRedTeam/FireRedTTS) or 'the unofficial implementation and pre-trained checkpoint' (SWivid/E2-TTS); Table 1 caption marks only DiTAR as '⧫ scores reported in DiTAR paper' (excluded)."
  },
  {
    "id": "speech-dmospeech2-report-bysa",
    "match": [
      "arxiv.org/html/2507.14988v1"
    ],
    "exact": true,
    "name": "DMOSpeech 2: Reinforcement Learning for Duration Prediction in Metric-Optimized Speech Synthesis (arXiv:2507.14988v1)",
    "kind": "paper",
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2507.14988v1",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-dmospeech2-report-bysa.txt",
    "attribution": "Yinghao Aaron Li et al.; 'DMOSpeech 2: Reinforcement Learning for Duration Prediction in Metric-Optimized Speech Synthesis'; arXiv:2507.14988v1, CC BY-SA 4.0. Changes: Table 2 Seed-TTS test-en/test-zh WER/CER and SIM for four re-run open baselines (CosyVoice 2, Spark-TTS, MaskGCT, LLaSA-8B); the authors' own F5-TTS teacher and DMOSpeech 2 rows excluded; error rates divided by 100. Share-alike: adapted rows remain CC BY-SA 4.0. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v1 abs page links http://creativecommons.org/licenses/by-sa/4.0/ and the HTML header reads 'License: CC BY-SA 4.0'. Table 2 caption: 'All samples from baseline models were synthesized using the official checkpoints released by the authors.'"
  },
  {
    "id": "speech-stylettszs-report-bysa",
    "match": [
      "arxiv.org/html/2409.10058v1"
    ],
    "exact": true,
    "name": "StyleTTS-ZS: Efficient High-Quality Zero-Shot Text-to-Speech Synthesis with Distilled Time-Varying Style Diffusion (arXiv:2409.10058v1)",
    "kind": "paper",
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2409.10058v1",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-stylettszs-report-bysa.txt",
    "attribution": "Yinghao Aaron Li et al.; 'StyleTTS-ZS: Efficient High-Quality Zero-Shot Text-to-Speech Synthesis with Distilled Time-Varying Style Diffusion'; arXiv:2409.10058v1, CC BY-SA 4.0. Changes: Table II MOS-N, MOS-S and WER for the three public LibriTTS baselines (StyleTTS 2 LibriTTS, HierSpeech++ LibriTTS-960, XTTS-v2); the authors' StyleTTS-ZS rows and Table I (samples obtained from other authors) excluded; MOS normalised (x-1)/4, WER divided by 100. Share-alike: adapted rows remain CC BY-SA 4.0. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v1 abs page links http://creativecommons.org/licenses/by-sa/4.0/ and the HTML header reads 'License: CC BY-SA 4.0'. Sec. IV-B: 'Official checkpoints trained on LibriTTS were used for all baseline models'; Appendix B-A: StyleTTS 2 'with the LibriTTS checkpoint at https://huggingface.co/yl4579/StyleTTS2-LibriTTS', HierSpeech++ 'with the LibriTTS-train-960 checkpoint', XTTSv2 'publicly available at https://huggingface.co/coqui/XTTS-v2 and we synthesized 40 samples'."
  },
  {
    "id": "speech-ditto-tts-report-ccby",
    "match": [
      "arxiv.org/html/2406.11427v2"
    ],
    "exact": true,
    "name": "DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors (arXiv:2406.11427v2)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2406.11427v2",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-ditto-tts-report-ccby.txt",
    "attribution": "Keon Lee et al. (KRAFTON); 'DiTTo-TTS: Diffusion Transformers for Scalable Text-to-Speech without Domain-Specific Factors'; arXiv:2406.11427v2, CC BY 4.0. Changes: English cross-sentence WER and SIM-o for re-run open checkpoints YourTTS (Table 2) and StyleTTS 2 / XTTS-v2 (Table 11); rows marked * (copied) and the authors' unreleased DiTTo models excluded; WER divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v2 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Appendix A.4: 'We use open-sourced checkpoints of YourTTS (github.com/Edresson/YourTTS), Simple-TTS, StyleTTS 2 (github.com/yl4579/StyleTTS2), XTTS-v2 (huggingface.co/coqui/XTTS-v2)'; for StyleTTS 2 and XTTS-v2 'For objective evaluation, we use samples generated from the model checkpoints'. Tables mark copied scores with an asterisk (excluded)."
  },
  {
    "id": "speech-voicecraft-report-bysa",
    "match": [
      "arxiv.org/html/2403.16973v3"
    ],
    "exact": true,
    "name": "VoiceCraft: Zero-Shot Speech Editing and Text-to-Speech in the Wild (arXiv:2403.16973v3)",
    "kind": "paper",
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2403.16973v3",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-voicecraft-report-bysa.txt",
    "attribution": "Puyuan Peng et al.; 'VoiceCraft: Zero-Shot Speech Editing and Text-to-Speech in the Wild'; arXiv:2403.16973v3, CC BY-SA 4.0. Changes: Table 6 zero-shot TTS WER, SIM and total naturalness MOS for VoiceCraft 830M, XTTS v2 and YourTTS; the authors' retrained VALL-E and FluentSpeech baselines excluded; WER divided by 100, MOS normalised (x-1)/4. Share-alike: adapted rows remain CC BY-SA 4.0. (v1/v2 of this paper are CC BY-NC-SA and are not used.) No endorsement.",
    "icon": "arxiv.org",
    "quote": "v3 abs page links http://creativecommons.org/licenses/by-sa/4.0/ and the HTML header reads 'License: CC BY-SA 4.0' (v1 and v2 are CC BY-NC-SA 4.0, so only v3 is cited). The zero-shot TTS set is the authors' own 250 prompt-transcript pairs from LibriTTS and GigaSpeech YouTube (Sec. 5.1), so every row was generated by the authors; XTTS v2 is cited to huggingface.co/coqui/XTTS-v2."
  },
  {
    "id": "speech-voxcpm-report-ccby",
    "match": [
      "arxiv.org/html/2509.24650v1"
    ],
    "exact": true,
    "name": "VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning (arXiv:2509.24650v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2509.24650v1",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-voxcpm-report-ccby.txt",
    "attribution": "OpenBMB / ModelBest VoxCPM team; 'VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning'; arXiv:2509.24650v1, CC BY 4.0. Changes: Table 5 listening-test N-MOS and S-MOS (ZH, EN) for VoxCPM, MaskGCT, CosyVoice 2 and IndexTTS 2, plus the authors' own VoxCPM Seed-TTS-eval row from Table 3; Table 3/4 baseline rows (mix of re-runs and copied values, not distinguishable) and the unreleased VoxCPM-Emilia excluded; MOS normalised (x-1)/4, error rates divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v1 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Sec. 4.1: 'Subjective evaluation involved Mean Opinion Score (MOS) tests rated by 20 native speakers on naturalness (N-MOS) and speaker similarity (S-MOS) using 5-point scales' — listening-test ratings require the authors to generate every system's audio. Objective baselines are 'obtained using official implementations with default settings, or as reported in their original papers' and are therefore not used."
  },
  {
    "id": "speech-spark-tts-report-ccby",
    "match": [
      "arxiv.org/html/2503.01710v1"
    ],
    "exact": true,
    "name": "Spark-TTS: An Efficient LLM-Based Text-to-Speech Model with Single-Stream Decoupled Speech Tokens (arXiv:2503.01710v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2503.01710v1",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-spark-tts-report-ccby.txt",
    "attribution": "Xinsheng Wang et al. (SparkAudio); 'Spark-TTS: An Efficient LLM-Based Text-to-Speech Model with Single-Stream Decoupled Speech Tokens'; arXiv:2503.01710v1, CC BY 4.0. Changes: Table 4 Spark-TTS own row (Seed-TTS test-zh CER/SIM, test-en WER/SIM); all baseline rows match values published in earlier papers and are excluded; error rates divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v1 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Only the authors' own Spark-TTS row is used; baseline rows in Table 4 reproduce values from the cited papers and are excluded."
  },
  {
    "id": "speech-raon-opentts-report-ccby",
    "match": [
      "arxiv.org/html/2605.20830v2"
    ],
    "exact": true,
    "name": "Raon-OpenTTS: Open Models and Data for Robust Text-to-Speech (arXiv:2605.20830v2)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2605.20830v2",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-raon-opentts-report-ccby.txt",
    "attribution": "KRAFTON Raon-OpenTTS team; 'Raon-OpenTTS: Open Models and Data for Robust Text-to-Speech'; arXiv:2605.20830v2, CC BY 4.0. Changes: Tables 1, 5, 6 and 8: Seed-TTS test-en WER/SIM, CV3-EN WER, CV3-Hard-EN WER/SIM/DNSMOS, Raon-OpenTTS-Eval overall WER/SIM and overall human SMOS for re-run open-weight baselines and the authors' released Raon-OpenTTS 0.3B/1B; closed-weight rows (Seed-TTS, CosyVoice 3 1.5B) reported from other papers, the unpinned F5-TTS (v0 vs v1) and Voxtral TTS rows, CMOS and per-condition splits excluded; WER divided by 100, MOS normalised (x-1)/4. No endorsement.",
    "icon": "arxiv.org",
    "quote": "v2 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Sec. 4.3: 'For all open-weight baseline models, we use their officially released checkpoints and follow the default or recommended inference settings provided by the authors, without additional tuning. For CosyVoice 3, we use the officially released 0.5B checkpoint, as the 1.5B variant is not available as open weights. For closed-weight models, we report the evaluation results as published by the original authors' (closed-weight rows excluded)."
  },
  {
    "id": "speech-rvcbench-ccby",
    "match": [
      "arxiv.org/html/2602.00443v2",
      "arxiv.org/html/2602.00443v3"
    ],
    "exact": true,
    "name": "RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models (arXiv:2602.00443v2)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2602.00443v2",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "license_url": "/licenses/speech-rvcbench-ccby.txt",
    "attribution": "Ruinan Jin, Xinting Liao, Hanlin Yu, Deval Pandya, Xiaoxiao Li; 'RVCBench: Benchmarking Robustness of Voice Cloning Across Modern Audio Generation Models'; arXiv:2602.00443v2, CC BY 4.0. Changes: selected LibriTTS clean-condition SIM and WER cells from Table 13 for eleven systems; SIM kept as raw cosine, WER kept as printed fraction. No endorsement.",
    "icon": "arxiv.org",
    "quote": "The v2 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. §3.1/Appendix C: the authors ran all 18 open-source voice-cloning models themselves through their toolkit (SpeechBrain ECAPA SIM, Whisper-medium WER, 1000-sample bootstrap); Table 13 is their own measurement. Model identities taken from the CC0-licensed RVCBench GitHub configs."
  },
  {
    "id": "speech-wayu-thai-tts-ccby",
    "match": [
      "arxiv.org/html/2609.03502v1"
    ],
    "exact": true,
    "name": "Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech (Wayu Research / Paxa Labs / Typhoon; arXiv:2609.03502v1)",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2609.03502v1",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-wayu-thai-tts-ccby.txt",
    "attribution": "Kunat Pipatanakul, Potsawee Manakul, Warit Sirichotedumrong, Sittipong Sripaisarnmongkol, Pakorn Nathong, Phatrasek Jirabovonvisut; 'Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech'; arXiv:2609.03502v1, CC BY 4.0. Changes: Table 8 Thai CER and Challenge-Set keyword accuracy for three systems, divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "quote": "The v1 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. The authors synthesised the evaluation audio with every system themselves ('We use only the transcripts and synthesize all evaluation audio', §3.1.1), including Gemini 3.1 Flash TTS via the gemini-3.1-flash-tts-preview API and the OmniVoice teacher."
  },
  {
    "id": "speech-cloneval-space-apache",
    "match": [
      "huggingface.co/spaces/amu-cai/Open_Voice_Cloning_Leaderboard/blob/f715ec805e912f43e79a5b948191ff044df27fec/results.csv"
    ],
    "exact": true,
    "name": "ClonEval · Open Voice Cloning Leaderboard (AMU)",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/amu-cai/Open_Voice_Cloning_Leaderboard/blob/f715ec805e912f43e79a5b948191ff044df27fec/README.md",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-cloneval-space-apache.txt",
    "attribution": "Open Voice Cloning Leaderboard / ClonEval, Iwona Christop, Tomasz Kuczyński and Marek Kubis (Adam Mickiewicz University), Hugging Face Space amu-cai/Open_Voice_Cloning_Leaderboard, Apache-2.0. Selected LibriSpeech test-clean WavLM similarity rows; catalogue IDs mapped. No endorsement.",
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: apache-2.0'; results.csv (all per-dataset/per-emotion WavLM and acoustic-feature similarities) is committed in the same Space repository and read locally by src/process_leaderboard_data.py (data_path='results.csv'). pages/about.md: the authors generate samples with each voice cloning model and average WavLM speaker-embedding cosine similarity. The ClonEval arXiv paper carries the default arXiv licence and is used only for identity."
  },
  {
    "id": "speech-arabic-tts-arena-apache",
    "match": [
      "huggingface.co/spaces/Navid-AI/Arabic-TTS-Arena/blob/6e003a592d4b5e64ce315fb635da22ec57f2fa82/leaderboard.json"
    ],
    "exact": true,
    "name": "Arabic TTS Arena (Navid AI) · leaderboard snapshot 2026-09-24",
    "kind": "leaderboard",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/spaces/Navid-AI/Arabic-TTS-Arena/blob/6e003a592d4b5e64ce315fb635da22ec57f2fa82/README.md",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-arabic-tts-arena-apache.txt",
    "attribution": "Arabic TTS Arena by Navid AI, Hugging Face Space Navid-AI/Arabic-TTS-Arena, Apache-2.0. Overall elo of active models from the pinned leaderboard.json; catalogue IDs mapped. No endorsement.",
    "icon": "huggingface.co",
    "quote": "Space card README.md at the pinned revision declares 'license: apache-2.0'; leaderboard.json (ratings, CIs, battle counts, per-dialect cells; last_updated 2026-09-24T00:01:34Z, total_votes 8647) is committed in the same Space repository (daily 'Upload leaderboard.json' commits) and read by app.py as LEADERBOARD_FILE."
  },
  {
    "id": "speech-balacoon-baselines-apache",
    "match": [
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/qwen3_tts/vctk/metrics.yaml",
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/qwen3_tts/daps_celeb/metrics.yaml",
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/sesame/vctk/metrics.yaml",
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/sesame/daps_celeb/metrics.yaml",
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/xtts/vctk/metrics.yaml",
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/xtts/daps_celeb/metrics.yaml",
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/zonos/vctk/metrics.yaml",
      "huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/zero-tts/zonos/daps_celeb/metrics.yaml"
    ],
    "exact": true,
    "name": "balacoon speech_gen_baselines · zero-shot TTS metrics (VCTK, DAPS-celeb)",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://huggingface.co/datasets/balacoon/speech_gen_baselines/blob/e14eac6aa6752aa68cb7b7ebfc379d51a1a085fd/README.md",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-balacoon-baselines-apache.txt",
    "attribution": "balacoon/speech_gen_baselines (Hugging Face dataset, Apache-2.0), evaluated with balacoon/speech_gen_eval (MIT) on balacoon/speech_gen_eval_testsets. Selected whisperv3_cer, utmos_mos and redimnet_secs values; UTMOS normalized (x-1)/4. No endorsement.",
    "icon": "huggingface.co",
    "quote": "Dataset card front matter: 'license: apache-2.0'; README: 'This dataset contains synthetic samples and evaluation results for various speech generation systems... Evaluation is done via speech_gen_eval.' metrics.yaml files are committed in the dataset. Generation notebooks in the MIT speech_gen_eval repo pin checkpoints (Qwen/Qwen3-TTS-12Hz-1.7B-Base, xtts_v2, Zyphra/Zonos-v0.1-transformer)."
  },
  {
    "id": "speech-amphion-maskgct-mit",
    "match": [
      "github.com/open-mmlab/Amphion/blob/26f6883110181f1dbfe95c70a7c7dbaf4de5f42a/models/tts/maskgct/README.md"
    ],
    "exact": true,
    "name": "Amphion MaskGCT README · zero-shot TTS results table",
    "kind": "first-party",
    "license": "MIT",
    "terms_url": "https://github.com/open-mmlab/Amphion/blob/26f6883110181f1dbfe95c70a7c7dbaf4de5f42a/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-amphion-maskgct-mit.txt",
    "attribution": "Copyright (c) 2023 Amphion. MaskGCT README (open-mmlab/Amphion), MIT License. Selected WER values for MaskGCT and the re-run XTTS-v2, VoiceCraft and CosyVoice baselines, percent divided by 100. No endorsement.",
    "icon": "github.com",
    "quote": "Repository root LICENSE is the MIT License ('Copyright (c) 2023 Amphion'), covering files in the repository including models/tts/maskgct/README.md. Provenance of rows from the MaskGCT paper (arXiv:2409.00750v3, default arXiv licence, identity only) Appendix D: XTTS-v2, VoiceCraft (830M_TTSEnhanced.pth) and CosyVoice (CosyVoice-300M) 'We use the official code and pre-trained checkpoint'; NaturalSpeech 3 and VoiceBox objective scores are copied ('We report the scores ... obtained from Ju et al.') and VALL-E is an Amphion reproduction — those rows are not imported."
  },
  {
    "id": "speech-ultraeval-audio-apache",
    "match": [
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/CosyVoice2.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/CosyVoice3.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/FireRedTTS3.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/IndexTTS2.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/VoxCPM.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/VoxCPM2.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/fishaudio-s2-pro.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/moss-tts-v1.5.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/qwen3_tts.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/MGM-Omni.md",
      "github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/replication/GLM-TTS.md"
    ],
    "exact": true,
    "name": "UltraEval-Audio (OpenBMB) · TTS replication results",
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/OpenBMB/UltraEval-Audio/blob/bead726925d43f526bed48a4a6a827595169429d/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "license_url": "/licenses/speech-ultraeval-audio-apache.txt",
    "attribution": "UltraEval-Audio by OpenBMB, Apache License 2.0. Selected reproduced Seed-TTS-Eval test-en WER / test-zh CER values from replication/*.md, percent divided by 100; official values in parentheses not used. No endorsement.",
    "icon": "github.com",
    "quote": "Repository root LICENSE is the Apache License, Version 2.0; the replication/*.md result documents are committed files of the repository. Each document states 'Performance format: reproduced_result(official_result)' — only the reproduced (first) value, measured by OpenBMB with the committed eval_cli commands, is used. VoxCPM/VoxCPM2 rows are OpenBMB's own models (by=lab)."
  },
  {
    "id": "speech-ttsds2-listening-mit",
    "match": [
      "huggingface.co/datasets/ttsds/listening_test/blob/4d4d98f452b17155d8bf9b44a62754407e21dd96/subjective_results.csv"
    ],
    "exact": true,
    "name": "TTSDS2 listening test results (ttsds/listening_test)",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://huggingface.co/datasets/ttsds/listening_test/blob/4d4d98f452b17155d8bf9b44a62754407e21dd96/README.md",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-ttsds2-listening-mit.txt",
    "attribution": "TTSDS2 listening test results by Christoph Minixhofer et al. (huggingface.co/datasets/ttsds/listening_test), MIT. Changes: per-system MOS and SMOS means over the Clean/Noisy/Wild/Kids domains computed from the published individual ratings and normalised (x-1)/4. No audio used. No endorsement.",
    "icon": "huggingface.co",
    "quote": "Dataset card front matter at the pinned revision: 'license: mit'; card text: 'This dataset contains all 11,000+ ratings collected for 20 synthetic speech systems for the TTSDS2 study ... The scores are MOS (Mean Opinion Score), CMOS (Comparative Mean Opinion Score) and SMOS (Speaker Similarity Mean Opinion Score).' The ratings file subjective_results.csv is part of that dataset. The accompanying paper (arXiv:2506.19441) is under the default arXiv licence and is used only to confirm system identity and that our aggregation reproduces its Table 2."
  },
  {
    "id": "speech-ttsds2-multilingual-mit",
    "match": [
      "huggingface.co/datasets/ttsds/leaderboard_data/blob/09ee7f1fb4cb32bb1ae187dfb4f6b7606e0ede69/ttsds/en_aggregated_multilingual.csv",
      "huggingface.co/datasets/ttsds/leaderboard_data/blob/09ee7f1fb4cb32bb1ae187dfb4f6b7606e0ede69/ttsds/zh_aggregated_multilingual.csv"
    ],
    "exact": true,
    "name": "TTSDS2 multilingual benchmark results (ttsds/leaderboard_data)",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://huggingface.co/datasets/ttsds/leaderboard_data/blob/09ee7f1fb4cb32bb1ae187dfb4f6b7606e0ede69/README.md",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-ttsds2-multilingual-mit.txt",
    "attribution": "TTSDS2 multilingual benchmark results by Christoph Minixhofer et al. (huggingface.co/datasets/ttsds/leaderboard_data), MIT. Changes: English and Chinese OVERALL scores selected and divided by 100. No audio used. No endorsement.",
    "icon": "huggingface.co",
    "quote": "Dataset card at the pinned revision is exactly '---\\nlicense: mit\\n---'; the result CSVs under ttsds/ are files of that dataset. Only the aggregated result files are used; the dataset's YouTube-derived audio is not downloaded or redistributed. The ttsdsbenchmark.com website and the ttsds/website repository (no licence) are not used."
  },
  {
    "id": "visual-geckonum",
    "match": [
      "arxiv.org/html/2406.14774v3"
    ],
    "exact": true,
    "name": "Evaluating Numerical Reasoning in Text-to-Image Models (GeckoNum, arXiv v3)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/visual-geckonum.txt",
    "attribution": "Ivana Kajić, Olivia Wiles, Isabela Albuquerque, Matthias Bauer, Su Wang, Jordi Pont-Tuset, Aida Nematzadeh (Google DeepMind). Evaluating Numerical Reasoning in Text-to-Image Models, arXiv:2406.14774v3 (6 Feb 2025), CC BY 4.0. Extracted Table 2 Task 1 human-annotated accuracy for selected publicly released models; labels mapped only to the stated model version. No values estimated or rerun. No endorsement; provided as-is.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2406.14774v3 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0).",
    "scope": "Only the authors' own human-annotated evaluation: Table 2 (S5.T2) Task 1 'Exact Number Generation' accuracy (five annotators per image, five seeds per prompt). Tasks 2-3, prompt-type breakdowns and auto-metric tables are not used. Withheld: Imagen-A/B/C/D and Muse-A/B (internal unreleased Google versions, lab), SD3 (no reference; API vs Medium not stated), Midjourney v6 (row first appears after v1 of June 2024; V6.0 vs V6.1 not stated)."
  },
  {
    "id": "video-worldreasonbench",
    "match": [
      "arxiv.org/html/2605.10434v1"
    ],
    "exact": true,
    "name": "WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (arXiv v1)",
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/video-worldreasonbench.txt",
    "attribution": "Keming Wu, Yijing Cui, Wenhan Xue, Qijie Wang, Xuan Luo, Zhiyuan Feng, Zuhao Yang, Sudong Wang, Sicong Jiang, Haowei Zhu, Zihan Wang, Ping Nie, Wenhu Chen, Bin Wang. WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors, arXiv:2605.10434v1 (11 May 2026), CC BY 4.0. Extracted the Table 4 Human Elo column for selected generators; model labels mapped only to the stated model/release. No values estimated or rerun. No endorsement; provided as-is.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2605.10434v1 links \"Rights to this article\" to http://creativecommons.org/licenses/by/4.0/ (CC BY 4.0). Appendix R also states the released WorldReasonBench/WorldRewardBench assets will be CC-BY 4.0; only numbers printed in the CC BY paper are used.",
    "scope": "Only the authors' own expert human preference ranking: Table 4 (S4.T4) 'Human Elo' (Bradley-Terry with Davidson ties over ~6K expert preference pairs from 15 trained annotators, WorldRewardBench). Judge Elo, QA accuracy and Score_PR columns (Qwen3.5-27B automatic metrics) are not used. Withheld: Kling (version/tier not stated), UniVideo (not in catalogue, one row), Sora2-12s (second duration of the same Sora 2 model; the 8-second setting is used)."
  },
  {
    "id": "paper-2403.19114v1",
    "name": "Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2403.19114v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2403.19114v1",
    "license_url": "/licenses/paper-2403.19114v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Chunqiu Steven Xia, Yinlin Deng, Lingming Zhang (2024). Top Leaderboard Ranking = Top Coding Proficiency, Always? EvoEval: Evolving Coding Benchmarks via LLM. arXiv:2403.19114v1. https://arxiv.org/abs/2403.19114v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2403.19114v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2 (pass@1 on HumanEval and the EvoEval overall column) with model identities from Table 5; the authors 'evaluate 51 popular state-of-the-art LLMs' themselves with greedy decoding (Section 4, Appendix A). Imported: HumanEval pass@1 and the printed EvoEval overall score. Not imported: bracketed HumanEval+ values (same generations re-scored), the five sub-benchmark columns separately, rows with unnamed API snapshots (gemini-1.0-pro), models outside the catalogue (text-bison-001, XwinCoder, Code Millenials, Speechless-CL, Magicoder-S-CL, Qwen-7B-Chat), Mistral-7B-v0.1 base (historical comparison set).",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2402.19173v1",
    "name": "StarCoder 2 and The Stack v2: The Next Generation — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2402.19173v1"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "terms_url": "https://arxiv.org/abs/2402.19173v1",
    "license_url": "/licenses/paper-2402.19173v1-CC-BY-SA-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Anton Lozhkov, Raymond Li, Loubna Ben Allal et al. (BigCode, 2024). StarCoder 2 and The Stack v2: The Next Generation. arXiv:2402.19173v1. https://arxiv.org/abs/2402.19173v1\nCC BY-SA 4.0: https://creativecommons.org/licenses/by-sa/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2402.19173v1 links the licence 'http://creativecommons.org/licenses/by-sa/4.0/' in its licence field, and the HTML version header states 'License: CC BY-SA 4.0'. Share-alike: rows adapted from this paper must be offered under CC BY-SA 4.0 (see adapted_data_file).",
    "scope": "Section 7 base-model comparisons run by the BigCode authors: Table 9 HumanEval and MBPP pass@1 (greedy, EvalPlus framework), Table 11 DS-1000 overall (completion, 40 samples, T=0.2), Table 14 GSM8K (PAL, 8-shot), Table 15 CRUXEval pass@1 for the StarCoder2 rows only. Not imported: HumanEval+/MBPP+ (same generations re-scored), MultiPL-E (no printed average), Table 13 CanItEdit and Table 16 FIM (captions say other models' results are taken from other papers), Table 15 baseline rows (identical to the CRUXEval authors' published values, so treated as copied).",
    "icon": "arxiv.org",
    "adapter_license": "CC-BY-SA-4.0",
    "publication_requirement": "Share-alike: retain attribution, change notice and licence when displaying or serving these rows; adapted rows must be offered under CC BY-SA 4.0.",
    "adapter_license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "adapted_data_file": "/data/llm-new-bysa.json"
  },
  {
    "id": "paper-2405.04324v1",
    "name": "Granite Code Models: A Family of Open Foundation Models for Code Intelligence — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2405.04324v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2405.04324v1",
    "license_url": "/licenses/paper-2405.04324v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Mayank Mishra, Matt Stallone, Gaoyuan Zhang et al. (IBM, 2024). Granite Code Models: A Family of Open Foundation Models for Code Intelligence. arXiv:2405.04324v1. https://arxiv.org/abs/2405.04324v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2405.04324v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Section 6: 'For all the benchmarks, we evaluate the baseline models (including ours) using the same script and environment for fair comparison.' Imported: Table 3 HumanEvalSynthesize 6-language average (base and instruct), Table 5 MBPP (base), Table 14 CRUXEval-I/O pass@1 (base), Table 15 MATH and GSM8K chain-of-thought (base). Not imported: MBPP+ (re-scoring), DS-1000 Table 7 (several baseline rows identical to the StarCoder 2 paper), Llama-3-70B base (authors flag a failed Python run inside the average), models outside the catalogue with a single evaluator (Mistral-7B-v0.2 base, Mixtral-8x22B base, Llama-3-8B base, OctoCoder, Llemma), other tables.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2308.12950v3",
    "name": "Code Llama: Open Foundation Models for Code — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2308.12950v3"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2308.12950v3",
    "license_url": "/licenses/paper-2308.12950v3-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Baptiste Rozière, Jonas Gehring, Fabian Gloeckle et al. (Meta, 2024). Code Llama: Open Foundation Models for Code. arXiv:2308.12950v3. https://arxiv.org/abs/2308.12950v3\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2308.12950v3 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 2 rows for Meta's own Code Llama, Code Llama - Python and Code Llama - Instruct 7B/13B/34B/70B: HumanEval pass@1 (zero-shot, greedy) and MBPP pass@1 (3-shot, greedy). Not imported: rows for other models (caption: 'Results for other models as provided by' other papers), Llama 2 base rows (historical comparison set / bridge caution), Unnatural Code Llama (unreleased).",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2406.11409v2",
    "name": "CodeGemma: Open Code Models Based on Gemma — authors' own runs",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2406.11409v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2406.11409v2",
    "license_url": "/licenses/paper-2406.11409v2-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "CodeGemma Team, Heri Zhao, Jeffrey Hui et al. (Google, 2024). CodeGemma: Open Code Models Based on Gemma. arXiv:2406.11409v2. https://arxiv.org/abs/2406.11409v2\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2406.11409v2 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field, and the HTML version header states 'License: CC BY 4.0'.",
    "scope": "Table 3 HumanEval and MBPP for Google's own CodeGemma 2B PT, 7B PT, 7B IT and Gemma 2B/7B PT (v1.0 checkpoints). Not imported: CodeGemma 1.1 rows (would be new single-evaluator models), Table 5 (collected from other papers), other tables.",
    "icon": "arxiv.org"
  },
  {
    "id": "arena-hard-auto-readme-v0-1",
    "name": "Arena-Hard-Auto v0.1 leaderboard (repository README, pre-v2)",
    "kind": "leaderboard",
    "match": [
      "github.com/lmarena/arena-hard-auto/blob/3071ef95d913bb5cd632641db336a3b6c7fe4bdd/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/lmarena/arena-hard-auto/blob/3071ef95d913bb5cd632641db336a3b6c7fe4bdd/LICENSE",
    "license_url": "/licenses/arena-hard-auto-v0-1-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "LMArena / Arena-Hard-Auto authors (Tianle Li, Wei-Lin Chiang et al.), github.com/lmarena/arena-hard-auto, Apache-2.0. Selected rows of the Arena-Hard-Auto v0.1 README leaderboard without style control; percentages converted to fractions; catalogue IDs mapped. No endorsement.",
    "quote": "Root LICENSE in the same pinned repository commit (3071ef95d9) is the Apache License 2.0; the v0.1 leaderboard tables are part of the repository README (documentation) with no separate licence. Same reuse basis as the already-approved arena-hard-auto-readme-v2 entry. The lmarena.ai / arena.ai websites are not used.",
    "scope": "README section 'Leaderboard' ('The following leaderboard has no style control', Updated: 11/14 = 2024-11-14): Arena-Hard-Auto v0.1 win rate vs gpt-4-0314, GPT-4-Turbo judge. The style-control table is not imported (same answers and judgments, different aggregation). Baseline gpt-4-0314 (50 by construction), unnamed snapshots and unreleased/system entries are not imported.",
    "icon": "github.com"
  },
  {
    "id": "cruxeval-published-results",
    "name": "CRUXEval published evaluation results (authors' generations and scores)",
    "kind": "dataset",
    "match": [
      "github.com/facebookresearch/cruxeval/blob/190faf16d175b5847b0af05d937872b1fb395942/samples/evaluation_results.zip"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://github.com/facebookresearch/cruxeval/blob/190faf16d175b5847b0af05d937872b1fb395942/LICENSE",
    "license_url": "/licenses/cruxeval-published-results-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Alex Gu, Baptiste Rozière, Hugh Leather, Armando Solar-Lezama, Gabriel Synnaeve, Sida I. Wang (Meta AI). CRUXEval, github.com/facebookresearch/cruxeval, MIT. Selected pass@1 fields from samples/evaluation_results.zip; percentages divided by 100; catalogue IDs mapped. No endorsement.",
    "quote": "README at commit 190faf16d1: 'CRUXEval is MIT licensed, as found in the LICENSE file.' and 'We also open-source generations and outputs for the models we display on the leaderboard ... To access the scored versions of the generations run unzip evaluation_results.zip.' The zip is committed in the same repository (added in the initial commit 10d71b4, 2024-01-03). The CRUXEval paper itself (arXiv 2401.03065) is under the default arXiv licence and is not used; the separate leaderboard website is not used.",
    "scope": "samples/evaluation_results.zip members *_temp0.2_input.json and *_temp0.2_output.json (direct prediction, no chain-of-thought), field pass_at_1. Model identities from inference/scripts/run_input_prediction.sh and openai/openai_run.py in the same commit. Not imported: temperature-0.8 / pass@5 and +cot members, codetulu-2-34b (not in the catalogue), Mistral-7B-v0.1 and phi-1.5 (historical comparison set).",
    "icon": "github.com"
  },
  {
    "id": "paper-2501.00656v3",
    "name": "2 OLMo 2 Furious (Team OLMo, Ai2) — authors' own evaluations of OLMo 2 Instruct and peer instruct models",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2501.00656v3"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2501.00656v3",
    "license_url": "/licenses/paper-2501.00656v3-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Team OLMo, Pete Walsh, Luca Soldaini, Dirk Groeneveld et al. (2024-2025). 2 OLMo 2 Furious. arXiv:2501.00656v3. https://arxiv.org/abs/2501.00656v3\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2501.00656v3 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field (evidence cached in the lane folder).",
    "scope": "Table 7 (OLMo 2 Instruct 1B/7B/13B/32B 'relative to peer open weight models'; caption: 'All models in this table are the instruction tuned variants'): columns AE2, BBH, GSM8K, IFE, MATH, MMLU, PQA, TQA for GPT-3.5 Turbo 0125, GPT-4o mini 0724, Gemma 3 1B/27B, Llama 3.2 1B, Qwen 2.5 1.5B/7B/14B/32B/72B, Ministral 8B 2410, Llama 3.1 8B/70B, Llama 3.3 70B, Tulu 3 8B, Gemma 2 9B/27B, QwQ 32B, SmolLM2 1.7B and OLMo 2 1B/7B/13B/32B. The paper evaluates these instruct models 'from the same evaluation suite used by Lambert et al. (2024)' (Tülu 3 / OLMES instruct suite, settings in Table 21) and describes running QwQ 32B itself with its recommended sampling. Average, FLOP, Safety and DROP columns (DROP is format-sensitive: Qwen 2.5 72B 34.2 vs 7B 54.4), the Mistral Small 24B row (version not stated), OLMo 1B 0724 / OLMo 7B 0424 rows, base-model tables (6, 9, 22) and ablation tables not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2502.02737v1",
    "name": "SmolLM2 paper (Hugging Face) — authors' own instruct-model evaluations",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2502.02737v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2502.02737v1",
    "license_url": "/licenses/paper-2502.02737v1-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "attribution": "Loubna Ben Allal, Anton Lozhkov, Elie Bakouch et al. (2025). SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model. arXiv:2502.02737v1. https://arxiv.org/abs/2502.02737v1\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names, model identities and settings clarified. No questions, tasks, model outputs or figures redistributed. No endorsement.",
    "quote": "The arXiv abstract page for 2502.02737v1 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field (evidence cached in the lane folder).",
    "scope": "Table 5 (1-2B instruction-tuned models: SmolLM2-1.7B-Instruct, Llama3.2-1B-Instruct, Qwen2.5-1.5B-Instruct): rows IFEval (Average), BBH (3-shot), HellaSwag, PIQA, GSM8K (5-shot), MATH (4-shot), HumanEval. §5.4: 'We evaluate the final instruct version of SmolLM2 and compare it with the instruct variants of Qwen2.5-1.5B and Llama3.2-1B'; evaluations use lighteval. MT-Bench, OpenRewrite-Eval, ARC and MMLU-Pro (lighteval MCF variant) rows, base-model and ablation tables not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "card2-ibm-granite-granite-3-3-8b-instruct",
    "name": "ibm-granite/granite-3.3-8b-instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ibm-granite/granite-3.3-8b-instruct/blob/51dd4bc2ade4059a6bd87649d68aa11e4fb2529b/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "/licenses/ibm-granite--granite-3.3-8b-instruct.txt",
    "terms_url": "https://huggingface.co/ibm-granite/granite-3.3-8b-instruct/blob/51dd4bc2ade4059a6bd87649d68aa11e4fb2529b/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'; no root LICENSE file; grant clause: 'each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work ...'. Selected own-lab result cells (Granite 3.1/3.2/3.3 2B and 8B Instruct rows of the 'Evaluation Results' tables) in the pinned README only; competitor rows (Llama-3.1-8B-Instruct, Qwen-2.5-7B-Instruct, DeepSeek-R1 distills), linked blogs and images are not used.",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "icon": "huggingface.co",
    "attribution": "ibm-granite. ibm-granite/granite-3.3-8b-instruct model card. Source: https://huggingface.co/ibm-granite/granite-3.3-8b-instruct/blob/51dd4bc2ade4059a6bd87649d68aa11e4fb2529b/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement."
  },
  {
    "id": "speech-stt-afrispeech200-report-ccby",
    "match": [
      "arxiv.org/html/2310.00274v1"
    ],
    "exact": true,
    "name": "AfriSpeech-200 (arXiv:2310.00274v1) · zero-shot WER of open ASR models on African-accented English",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2310.00274v1",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-stt-afrispeech200-report-ccby.txt",
    "attribution": "Tobi Olatunji, Tejumade Afonja, Aditya Yadavalli, Chris Chinenye Emezue, Sahib Singh, Bonaventure F. P. Dossou, Joanne Osuchukwu, Salomey Osei, Atnafu Lambebo Tonja, Naome Etori, Clinton Mbataku, 'AfriSpeech-200: Pan-African Accented Speech Dataset for Clinical and General Domain ASR', arXiv:2310.00274v1 (2023), CC BY 4.0. Table 4 test-set WER of open-source checkpoints transcribed by script. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2310.00274v1 links http://creativecommons.org/licenses/by/4.0/ and the HTML states 'License: CC BY 4.0' (the AfriSpeech-200 dataset itself is CC BY-NC-SA 4.0; only the paper's numbers are used). §4: the authors 'evaluate multiple SOTA pre-trained models' on the new benchmark; Table 4 lists each HF checkpoint with its own LibriSpeech and AfriSpeech runs, and 'Missing values indicate incomplete or failed experiments' — the authors' own runs. Commercial API rows are unversioned and not used."
  },
  {
    "id": "speech-stt-liteasr-report-ccby",
    "match": [
      "arxiv.org/html/2502.20583v2"
    ],
    "exact": true,
    "name": "LiteASR (arXiv:2502.20583v2) · ESB WER of Whisper and LiteASR-compressed encoders",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2502.20583v2",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-stt-liteasr-report-ccby.txt",
    "attribution": "Keisuke Kamahori, Jungo Kasai, Noriyuki Kojima, Baris Kasikci, 'LiteASR: Efficient Automatic Speech Recognition with Low-Rank Approximation', arXiv:2502.20583v2, CC BY 4.0. Table 1 ESB average WER transcribed by script, percent divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "quote": "arXiv abstract page for 2502.20583v2 links http://creativecommons.org/licenses/by/4.0/ and the HTML states 'License: CC BY 4.0'. Table 1 reports the authors' own ESB evaluation of the original Whisper large-v3, large-v3-turbo and medium checkpoints and of their compressed variants (released as efficient-speech/lite-whisper-*); the values differ from leaderboard numbers, so they are not copied."
  },
  {
    "id": "speech-zstts-eval-casanova-mit",
    "match": [
      "github.com/Edresson/ZS-TTS-Evaluation/blob/a9086e4192b6ccba9f06c492a5ef68b52ae8ccbe/Evaluation/XTTS_v2.0.2/metrics.csv",
      "github.com/Edresson/ZS-TTS-Evaluation/blob/a9086e4192b6ccba9f06c492a5ef68b52ae8ccbe/Evaluation/YourTTS_original/metrics.csv",
      "github.com/Edresson/ZS-TTS-Evaluation/blob/a9086e4192b6ccba9f06c492a5ef68b52ae8ccbe/Evaluation/Tortoise_num_autoregressive_samples_256_diffusion_iterations_200/metrics.csv",
      "github.com/Edresson/ZS-TTS-Evaluation/blob/a9086e4192b6ccba9f06c492a5ef68b52ae8ccbe/Evaluation/StyleTTS2/metrics.csv"
    ],
    "exact": true,
    "name": "ZS-TTS-Evaluation (Edresson Casanova) · zero-shot TTS metrics of XTTS, YourTTS, TorToiSe and StyleTTS 2",
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/Edresson/ZS-TTS-Evaluation/blob/a9086e4192b6ccba9f06c492a5ef68b52ae8ccbe/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-28",
    "license_url": "/licenses/speech-zstts-eval-casanova-mit.txt",
    "attribution": "ZS-TTS-Evaluation, Copyright (c) 2024 Edresson Casanova, MIT (https://github.com/Edresson/ZS-TTS-Evaluation). English CER (divided by 100) and ECAPA2 SECS transcribed by script from the committed metrics.csv files. No endorsement.",
    "icon": "github.com",
    "quote": "Repository root LICENSE at a9086e4: 'MIT License / Copyright (c) 2024 Edresson Casanova / Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction'; the metrics.csv files are committed repository files produced by the repository's eval_TTS.py over each system's committed generated audio (README: 'This repository gathers our efforts to evaluate/compare the current Zero-shot Multi-speaker TTS (ZS-TTS) systems using objective metrics'). Only the numerical metrics are reused; no audio."
  },
  {
    "id": "speech-espnet-librispeech-20260929",
    "name": "ESPnet LibriSpeech ASR recipe results (pinned September 2026)",
    "match": [
      "github.com/espnet/espnet/blob/b90237261607e61f5e80b4bb8c362f80b209da3c/egs2/librispeech/asr1/README.md",
      "github.com/espnet/espnet/blob/b90237261607e61f5e80b4bb8c362f80b209da3c/egs2/owsm_v3.1/s2t1/README.md"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/espnet/espnet/blob/b90237261607e61f5e80b4bb8c362f80b209da3c/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "license_url": "/licenses/speech-espnet-librispeech-20260929.txt",
    "icon": "github.com",
    "quote": "Apache-2.0 covers documentation source. Only the ESPnet recipe authors’ own WER tables, not third-party comparison numbers or external model artefacts. Test-clean and test-other WER only; CER/TER and development-set results are omitted. Decoder/LM choices preserved in each model identity and result note.",
    "scope": "Apache-2.0 covers documentation source. Only the ESPnet recipe authors’ own WER tables, not third-party comparison numbers or external model artefacts. Also ESPnet’s own Whisper-medium run in the OWSM v3.1 README at the same commit (greedy decoding, Whisper normaliser; arXiv 2309.13876: “we prepare all test data in ESPnet and evaluate Whisper in the same setup”). Test-clean and test-other WER only; CER/TER and development-set results are omitted. Decoder/LM choices preserved in each model identity and result note.",
    "attribution": "ESPnet developers and recipe contributors. Extracted selected test WER cells from the committed LibriSpeech recipe README; percentages divided by 100. No endorsement."
  },
  {
    "id": "speech-nvidia-dle-asr",
    "name": "NVIDIA DeepLearningExamples Jasper and QuartzNet READMEs",
    "match": [
      "github.com/NVIDIA/DeepLearningExamples/blob/729963dd47e7c8bd462ad10bfac7a7b0b604e6dd/PyTorch/SpeechRecognition/Jasper/README.md",
      "github.com/NVIDIA/DeepLearningExamples/blob/729963dd47e7c8bd462ad10bfac7a7b0b604e6dd/PyTorch/SpeechRecognition/QuartzNet/README.md"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/NVIDIA/DeepLearningExamples/blob/729963dd47e7c8bd462ad10bfac7a7b0b604e6dd/PyTorch/SpeechRecognition/Jasper/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "license_url": "/licenses/speech-nvidia-dle-asr.txt",
    "icon": "github.com",
    "quote": "Apache-2.0 covers the documentation source. Both READMEs sit in the same repository at the same commit and report the same protocol (NVIDIA's own A100 training-accuracy table, greedy decoding, LibriSpeech test-clean / test-other), so they are one evaluator's run. Only those tables; not other seeds, hardware throughput, or third-party paper comparisons.",
    "scope": "Apache-2.0 covers the documentation source. Only NVIDIA’s own A100 training-accuracy table (greedy decoding); not other seeds, hardware throughput, or third-party paper comparisons.",
    "attribution": "NVIDIA Corporation. Selected LibriSpeech test-clean/test-other WER cells; percentage / 100. No endorsement."
  },
  {
    "id": "speech-vits-icml2021",
    "name": "VITS: ICML 2021 original listening studies",
    "match": [
      "proceedings.mlr.press/v139/kim21f/kim21f.pdf"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "CC-BY-4.0",
    "terms_url": "https://proceedings.mlr.press/pmlr-license-agreement.html",
    "verdict": "allowed",
    "checked": "2026-09-29",
    "license_url": "/licenses/speech-vits-icml2021.txt",
    "icon": "github.com",
    "quote": "PMLR publication agreement paragraph 2 grants CC BY 4.0 in the article and paragraph 3 requires the PMLR publication link. Scope is the PMLR published article, not the separately licensed arXiv upload. Tables 1 and 3 are the authors’ own crowd-sourced listening studies (§3.4, §4.1, §4.2).",
    "scope": "PMLR publication agreement paragraph 2 grants CC BY 4.0 in the article and paragraph 3 requires the PMLR publication link. Scope is the PMLR published article, not the separately licensed arXiv upload. Tables 1 and 3 are the authors’ own crowd-sourced listening studies (§3.4, §4.1, §4.2).",
    "attribution": "Jaehyeon Kim, Jungil Kong and Juhee Son. Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech. Proceedings of ICML, PMLR 139:5530–5540, 2021. https://proceedings.mlr.press/v139/kim21f.html . Changes: selected MOS means, stored as (MOS − 1) / 4; 95% CIs preserved in result notes. No endorsement."
  },
  {
    "id": "visual-evalcrafter-2023",
    "name": "EvalCrafter: original 2023 text-to-video evaluations",
    "match": [
      "arxiv.org/html/2310.11440v3"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://arxiv.org/abs/2310.11440v3",
    "verdict": "allowed",
    "checked": "2026-09-29",
    "license_url": "/licenses/visual-evalcrafter-2023.txt",
    "icon": "arxiv.org",
    "quote": "The v3 arXiv abstract links CC BY 4.0. §3–5 and Appendix B describe the authors’ own generations and evaluations. Table 2 reports four human-preference-aligned composite scores; these are not direct human MOS or percentages of tasks passed.",
    "scope": "Table 2 four aspect composites only; 17 constituent metrics and correlation tables omitted to avoid duplicating the same evidence. Exact 2023 API versions and checkpoint/pipeline identities retained.",
    "attribution": "Yaofang Liu, Xiaodong Cun, Xuebo Liu, Xintao Wang, Yong Zhang, Haoxin Chen, Yang Liu, Tieyong Zeng, Raymond Chan and Ying Shan. EvalCrafter: Benchmarking and Evaluating Large Video Generation Models, arXiv:2310.11440v3, CC BY 4.0. Selected Table 2 values /100; versioned systems and evaluation settings retained. No endorsement."
  },
  {
    "id": "speech-nvidia-fastpitch-readme",
    "name": "NVIDIA FastPitch: original listening studies in committed README",
    "match": [
      "github.com/NVIDIA/DeepLearningExamples/blob/729963dd47e7c8bd462ad10bfac7a7b0b604e6dd/PyTorch/SpeechSynthesis/FastPitch/README.md"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "BSD-3-Clause",
    "terms_url": "https://github.com/NVIDIA/DeepLearningExamples/blob/729963dd47e7c8bd462ad10bfac7a7b0b604e6dd/PyTorch/SpeechSynthesis/FastPitch/LICENSE",
    "verdict": "allowed",
    "checked": "2026-09-29",
    "license_url": "/licenses/speech-nvidia-fastpitch-readme.txt",
    "icon": "github.com",
    "quote": "BSD 3-Clause permits redistribution with notices. Scope is the committed README’s own NVIDIA research MOS and pairwise preference results; no external service, copied third-party comparison data or model weights are licensed by this entry.",
    "attribution": "NVIDIA Corporation; Adrian Łańcucki (FastPitch 2020), and NVIDIA FastPitch 1.1 authors. Selected README listening-study means, MOS stored as (MOS−1)/4; preference fractions unchanged. Original papers: arXiv:2006.06873 and arXiv:2108.10447. No endorsement."
  },
  {
    "id": "speech-gradtts-icml2021",
    "name": "Grad-TTS: original ICML 2021 listening study",
    "match": [
      "proceedings.mlr.press/v139/popov21a/popov21a.pdf"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC-BY-4.0",
    "terms_url": "https://proceedings.mlr.press/pmlr-license-agreement.html",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "license_url": "/licenses/speech-gradtts-icml2021.txt",
    "icon": "arxiv.org",
    "quote": "PMLR publication agreement paragraph 2 grants CC BY 4.0 in the published article; paragraph 3 requires its publication link. §4.1 describes original listening evaluations for all systems. Scope: PMLR paper, not the arXiv version.",
    "scope": "Original Table 2 listening-study systems with HiFi-GAN. Grad-TTS-10, xcmyz FastSpeech, and the study-specific Glow-TTS and Tacotron2 + HiFi-GAN systems are separate identities. No score assigned to Coqui or NVIDIA WaveGlow distributions. No extra Grad-TTS iteration settings or human ground truth counted as additional models.",
    "attribution": "Vadim Popov, Ivan Vovk, Vladimir Gogoryan, Tasnima Sadekova and Mikhail Kudinov. Grad-TTS: A Diffusion Probabilistic Model for Text-to-Speech. ICML 2021, PMLR 139. https://proceedings.mlr.press/v139/popov21a.html . Table 2 selected means transformed as (MOS−1)/4. No endorsement. Added the two original baseline systems as explicitly study-specific comparison entries, without claiming exact downloadable checkpoint equivalence."
  },
  {
    "id": "speech-versa-toolkit-ccby",
    "name": "VERSA toolkit paper: TTS demonstration (arXiv:2412.17667v2)",
    "match": [
      "arxiv.org/html/2412.17667v2"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2412.17667v2",
    "license_url": "/licenses/speech-versa-toolkit-ccby.txt",
    "verdict": "allowed",
    "checked": "2026-10-05",
    "icon": "arxiv.org",
    "quote": "The v2 abs page links http://creativecommons.org/licenses/by/4.0/ and the HTML header reads 'License: CC BY 4.0'. Appendix C.2: 'we utilize a range of open-source TTS models for the TTS evaluation ... we use the common LibriSpeech test-clean set for evaluation ... Inference is performed using the open-source pipelines provided by the model developers, with no additional hyperparameter tuning.' Table 6 is the authors' own VERSA evaluation of those generations.",
    "scope": "Appendix C.2 Table 6 UTMOS column for the six systems whose Table 5 checkpoint links match catalogue entries (MeloTTS-English, Parler-TTS Mini v1, CosyVoice-300M, Plachtaa VALL-E X, Amphion VALL-E (amphion/valle), Amphion NaturalSpeech 2). The Whisper-WER column (Whisper size not stated) was held back on 2026-10-05 because it fitted poorly (series noise 0.72; TTS leave-one-out r without floor-weight series fell from 0.658 to 0.629 and the error-metric p90 rose from 6.8 to 9.5 points). Other columns are not imported, to avoid counting correlated metrics of the same generations several times; ESPnet-TTS, ESPnet-SpeechLM, ChatTTS, EmotiVoice and WhisperSpeech small are not catalogue models.",
    "attribution": "Jiatong Shi, Hye-jin Shim, Jinchuan Tian, Siddhant Arora, Haibin Wu, Darius Petermann, Jia Qi Yip, You Zhang, Yuxun Tang, Wangyou Zhang, Dareen Safar Alharthi, Yichen Huang, Koichi Saito, Jionghao Han, Yiwen Zhao, Chris Donahue and Shinji Watanabe. VERSA: A Versatile Evaluation Toolkit for Speech, Audio, and Music, arXiv:2412.17667v2, CC BY 4.0. Changes: selected Table 6 UTMOS values stored as (x - 1) / 4. No endorsement."
  },
  {
    "id": "report-2512.13607v2",
    "name": "NVIDIA — Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2512.13607v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2512.13607v2",
    "quote": "arXiv abstract page for 2512.13607v2 (v2 submitted Fri, 27 Mar 2026): licence link <a href=\"http://creativecommons.org/licenses/by/4.0/\" title=\"Rights to this article\" class=\"has_license\"> with the by-4.0 licence icon (\"view license\"). Only Table 9 LiveCodeBench Pro 25Q2 cells are selected: NVIDIA's own Nemotron-Nano-9B-v2 row, plus baseline rows flagged needs_review because the report uses official numbers where available.",
    "attribution": "NVIDIA. Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models. arXiv:2512.13607v2. CC BY 4.0. Source: https://arxiv.org/html/2512.13607v2. Changes: selected numerical cells transcribed into JSON; percentages divided by 100; benchmark names and settings clarified. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "first-party-01ai",
    "name": "01.AI official model cards and API documentation",
    "kind": "first-party",
    "match": [
      "huggingface.co/01-ai/Yi-34B-Chat/blob/cf02cb50f2a03dead2fe205766a1c5598a90bf80/README.md",
      "huggingface.co/01-ai/Yi-Coder-9B-Chat/blob/356a1f8d4e4a606d0b879e54191ca809918576b8/README.md",
      "platform.lingyiwanwu.com/docs",
      "huggingface.co/01-ai/Yi-1.5-34B-Chat/blob/fa4ffba162f20948bf77c2a30eca952bf0812b7f/config.json",
      "huggingface.co/01-ai/Yi-1.5-6B-Chat/blob/771924d1c83d67527d665913415d7086f11ea9c0/config.json",
      "huggingface.co/01-ai/Yi-1.5-9B-Chat/blob/1a0fc698cf883c4f5c325f026ca79f0ebd9955a5/config.json",
      "huggingface.co/01-ai/Yi-1.5-34B-Chat-16K/blob/ff74452e11f0f749ab872dc19b1dd3813c25c4d8/config.json",
      "huggingface.co/01-ai/Yi-1.5-34B/blob/58a29f6dca2a4eda38821c086c29c1906114aabb/config.json",
      "huggingface.co/01-ai/Yi-1.5-6B/blob/157a3d77ab5a8a8f079510eaa25169b547481f06/config.json",
      "huggingface.co/01-ai/Yi-1.5-9B-Chat-16K/blob/2f8753bed95bd4fede3fd64f8305428620b4b212/config.json",
      "huggingface.co/01-ai/Yi-1.5-9B/blob/80d5471b1eae28beae33e06eadbd4b48e74d4ce1/config.json",
      "huggingface.co/01-ai/Yi-34B-200K/blob/e0ae1afac6b69f604556efd441ab4befafb2a835/config.json",
      "huggingface.co/01-ai/Yi-6B-Chat/blob/2dbf63b0cb7bc493c0243502c6e6111a36e3a093/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/01-ai/Yi-34B-Chat/blob/cf02cb50f2a03dead2fe205766a1c5598a90bf80/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "01.AI official model cards and API documentation, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-ai2",
    "name": "Ai2 (Allen Institute for AI) official technical reports",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/2501.00656v3",
      "arxiv.org/abs/2512.13961v2",
      "huggingface.co/allenai/OLMo-1B-hf/blob/aee7752d9c08ee4775e9b0091426d8410e8f6a89/README.md",
      "huggingface.co/allenai/OLMo-7B-Twin-2T-hf/blob/838af31020e1ebe5d627c9888e090b1fe4d2576e/README.md",
      "huggingface.co/allenai/OLMo-7B-hf/blob/11fb3186a2e4f681edea621fa8b4345147a9db6a/README.md",
      "huggingface.co/allenai/Llama-3.1-Tulu-3-70B-DPO/blob/2dab5183da9252bedf6039a0b09c86b9ca4d0ba9/config.json",
      "huggingface.co/allenai/Llama-3.1-Tulu-3-8B/blob/666943798adbde0b1aff34626007e26986a3c107/config.json",
      "huggingface.co/allenai/Llama-3.1-Tulu-3-70B/blob/cfc1d855e534a0b9b82a9cea6bf9e8dda30b10d7/config.json",
      "huggingface.co/allenai/OLMo-2-0425-1B-Instruct/blob/48d788eca847d4d7548f375ad03d3c9312f6139e/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2501.00656v3",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Ai2 (Allen Institute for AI) official technical reports, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-07"
  },
  {
    "id": "first-party-arcee",
    "name": "Arcee AI official model cards",
    "kind": "first-party",
    "match": [
      "huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/dc6a99b6b74202880e31491a445a94884c97890c/README.md",
      "docs.arcee.ai/get-started/pricing.md",
      "huggingface.co/arcee-ai/AFM-4.5B/blob/e84d19711eca5817c1608cb344d931bf05c966dd/config.json",
      "huggingface.co/arcee-ai/Trinity-Mini/blob/cdb81e8130815fec299cccc8f8700004b8366e1e/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/dc6a99b6b74202880e31491a445a94884c97890c/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Arcee AI official model cards, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-bigcode",
    "name": "BigCode official model cards",
    "kind": "first-party",
    "match": [
      "huggingface.co/bigcode/starcoder2-15b/blob/46d44742909c03ac8cee08eb03fdebce02e193ec/README.md",
      "huggingface.co/bigcode/starcoder2-3b/blob/733247c55e3f73af49ce8e9c7949bf14af205928/README.md",
      "huggingface.co/bigcode/starcoder2-7b/blob/bb9afde76d7945da5745592525db122d4d729eb1/README.md",
      "huggingface.co/bigcode/starcoder2-15b-instruct-v0.1/blob/ffb8dd9776ba9a66d655ecd962e882f3013e9f7c/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/bigcode/starcoder2-15b/blob/46d44742909c03ac8cee08eb03fdebce02e193ec/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "BigCode official model cards, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-bigscience",
    "name": "BigScience official model cards",
    "kind": "first-party",
    "match": [
      "huggingface.co/bigscience/bloom-1b1/blob/eb3dd7399312f5f94fd13f41d2f318117d3eb1e4/README.md",
      "huggingface.co/bigscience/bloom-1b7/blob/cc72a88036c2fb937d65efeacc57a0c2ef5d6fe5/README.md",
      "huggingface.co/bigscience/bloom-3b/blob/52bc5b43010b4844513826b8be3f78c7344c37d7/README.md",
      "huggingface.co/bigscience/bloom-560m/blob/ac2ae5fab2ce3f9f40dc79b5ca9f637430d24971/README.md",
      "huggingface.co/bigscience/bloom-7b1/blob/6232703e399354503377bf59dfbb8397fd569e4a/README.md",
      "huggingface.co/bigscience/bloom/blob/7f10a99ce7c08f03c7719a586cb2cbda1433ac05/README.md"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/bigscience/bloom-1b1/blob/eb3dd7399312f5f94fd13f41d2f318117d3eb1e4/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "BigScience official model cards, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "first-party-databricks",
    "name": "Databricks official DBRX announcement",
    "kind": "first-party",
    "match": [
      "databricks.com/blog/introducing-dbrx-new-state-art-open-llm"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Databricks official DBRX announcement, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "first-party-deep-cogito",
    "name": "Deep Cogito official model cards",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepcogito/cogito-v1-preview-llama-3B/blob/5a0ad6f39500e9af65e727f15d5554352933ed65/README.md",
      "huggingface.co/deepcogito/cogito-v1-preview-llama-8B/blob/64c42369b3f322fbffb277bfff146551dd2823cc/README.md",
      "huggingface.co/deepcogito/cogito-v1-preview-qwen-14B/blob/b5a6c9ab8fc8e5bc786ae88f66af479096367566/README.md",
      "huggingface.co/deepcogito/cogito-v1-preview-qwen-32B/blob/118e6e43c46a51667f544dae9cbe4027f59eb697/README.md",
      "huggingface.co/deepcogito/cogito-v1-preview-llama-70B/blob/1d624e2293b5b35f9cfd2349f8e02c7ebf32ca83/README.md",
      "huggingface.co/deepcogito/cogito-v2-preview-llama-70B/blob/1e1d12e8eaebd6084a8dcf45ecdeaa2f4b8879ce/README.md"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/deepcogito/cogito-v1-preview-llama-3B/blob/5a0ad6f39500e9af65e727f15d5554352933ed65/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Deep Cogito official model cards, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-eleutherai",
    "name": "EleutherAI official model cards and technical reports",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/2304.01373v2",
      "huggingface.co/EleutherAI/gpt-neox-20b/blob/c292233c833e336628618a88a648727eb3dff0a7/README.md",
      "huggingface.co/EleutherAI/gpt-neo-125m/blob/21def0189f5705e2521767faed922f1f15e7d7db/config.json",
      "huggingface.co/EleutherAI/gpt-neo-1.3B/blob/dbe59a7f4a88d01d1ba9798d78dbe3fe038792c8/config.json",
      "huggingface.co/EleutherAI/gpt-neo-2.7B/blob/e24fa291132763e59f4a5422741b424fb5d59056/config.json",
      "huggingface.co/EleutherAI/pythia-2.8b-deduped/blob/7d977fed8c4ce9649816af8cd5fe36a639cbe5b2/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2304.01373v2",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "EleutherAI official model cards and technical reports, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-google-models",
    "name": "Google / Google DeepMind official open-model cards and technical reports",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/1910.10683v4",
      "arxiv.org/abs/2403.08295v4",
      "arxiv.org/abs/2408.00118v3",
      "huggingface.co/google/gemma-2b/blob/9cf48e52b224239de00d483ec8eb84fb8d0f3a3a/README.md",
      "huggingface.co/google/gemma-7b/blob/ff6768d9368919a1f025a54f9f5aa0ee591730bb/README.md"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/1910.10683v4",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Google / Google DeepMind official open-model cards and technical reports, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "first-party-huggingface",
    "name": "Hugging Face (SmolLM) official technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/2502.02737v1",
      "huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1/blob/a3be084543d278e61b64cd600f28157afc79ffd3/config.json",
      "huggingface.co/HuggingFaceH4/starchat2-15b-v0.1/blob/837a7b6ac3df5f91de138e3592258f93c80d93d3/config.json",
      "huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct/blob/a10cc1512eabd3dde888204e902eca88bddb4951/config.json",
      "huggingface.co/HuggingFaceTB/SmolLM2-135M-Instruct/blob/12fd25f77366fa6b3b4b768ec3050bf629380bac/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2502.02737v1",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Hugging Face (SmolLM) official technical report, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-lmsys",
    "name": "LMSYS official FastChat repository documentation",
    "kind": "first-party",
    "match": [
      "github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/README.md",
      "huggingface.co/lmsys/vicuna-33b-v1.3/blob/ef8d6becf883fb3ce52e3706885f761819477ab4/config.json",
      "huggingface.co/lmsys/vicuna-13b-v1.3/blob/6566e9cb1787585d1147dcf4f9bc48f29e1328d2/config.json",
      "huggingface.co/lmsys/vicuna-7b-v1.3/blob/236eeeab96f0dc2e463f2bebb7bb49809279c6d6/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "LMSYS official FastChat repository documentation, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-microsoft",
    "name": "Microsoft official model cards and technical reports",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/2306.11644v2",
      "arxiv.org/abs/2309.05463v1",
      "huggingface.co/microsoft/Phi-3-medium-128k-instruct/blob/a088b37c71d441ab6d862bb3fcfe6165b3014702/README.md",
      "huggingface.co/microsoft/Phi-3-medium-4k-instruct/blob/48d87cd5e0523b430d77e93becd3655cd6897230/README.md",
      "huggingface.co/microsoft/Phi-3.5-MoE-instruct/blob/43688451b462a3351d8580625ebe1931adb3986d/README.md",
      "huggingface.co/microsoft/Phi-3.5-mini-instruct/blob/2fe192450127e6a83f7441aef6e3ca586c338b77/README.md",
      "huggingface.co/microsoft/phi-2/blob/810d367871c1d460086d9f82db8696f2e0a0fcd0/README.md",
      "huggingface.co/microsoft/Orca-2-13b/blob/2539ff53e6baa4cc603774ad5a2d646f4041ea4e/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2306.11644v2",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Microsoft official model cards and technical reports, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-nous-research",
    "name": "Nous Research official technical reports",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/2408.11857v1",
      "huggingface.co/NousResearch/Hermes-3-Llama-3.2-3B/blob/7f1a6bec8cdce6551014fd5bbeb4cd8c0f1fbeab/config.json",
      "huggingface.co/NousResearch/Hermes-2-Theta-Llama-3-70B/blob/8948247fa83f3f970ad24b3394d5656507642922/config.json",
      "huggingface.co/NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO/blob/286ae6737d048ad1d965c2e830864df02db50f2f/config.json",
      "huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B/blob/f798274b30e7d2d4797c369edcc0cc7473b6e6f2/config.json",
      "huggingface.co/NousResearch/Hermes-2-Pro-Mistral-7B/blob/24dbda51d986efa05201eec7a093ac91fc5519cf/config.json",
      "huggingface.co/NousResearch/Hermes-2-Theta-Llama-3-8B/blob/57a73110702e7b05ba3f39fef36297454c680725/config.json",
      "huggingface.co/NousResearch/Nous-Hermes-13b/blob/24e8c03148ffd1f3e469744dfc24ad2ad82848f8/config.json",
      "huggingface.co/NousResearch/Nous-Hermes-2-SOLAR-10.7B/blob/14c1fbe2f71acdcd58247b30d5439bd572d52386/config.json",
      "huggingface.co/NousResearch/Nous-Hermes-2-Yi-34B/blob/fcb0a8847e76aea14aba9aa44009d4418ad7c18f/config.json",
      "huggingface.co/NousResearch/Nous-Hermes-2-Mistral-7B-DPO/blob/ebec0a691037d38955727d6949798429a63929dd/config.json",
      "huggingface.co/NousResearch/Hermes-3-Llama-3.1-70B/blob/1fc86da0ce9cdb14cd775ad270bc7d1b4bf70ede/config.json",
      "huggingface.co/NousResearch/Hermes-3-Llama-3.1-405B/blob/b9fe8ede1f5075f171b2c0a235e20ae4aba33767/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2408.11857v1",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Nous Research official technical reports, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-nvidia",
    "name": "NVIDIA official model cards",
    "kind": "first-party",
    "match": [
      "huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF/blob/031d4042f36adc1a52cca51b331d25cbe3cf1022/README.md",
      "huggingface.co/nvidia/Mistral-NeMo-Minitron-8B-Instruct/blob/64261e01bfabab557842e5803a7ea96cc09489a5/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF/blob/031d4042f36adc1a52cca51b331d25cbe3cf1022/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "NVIDIA official model cards, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-openchat",
    "name": "OpenChat official model cards",
    "kind": "first-party",
    "match": [
      "huggingface.co/openchat/openchat-3.5-0106/blob/ff058fda49726ecf4ea53dc1635f917cdb8ba36b/README.md",
      "huggingface.co/openchat/openchat_3.5/blob/0fc98e324280bc4bf5d2c30ecf7b97b84fb8a19b/README.md",
      "huggingface.co/openchat/openchat-3.5-0106-gemma/blob/5316738a9049ab32e3b9a49eb0e9dfb49de8a725/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/openchat/openchat-3.5-0106/blob/ff058fda49726ecf4ea53dc1635f917cdb8ba36b/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "OpenChat official model cards, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-qwen",
    "name": "Qwen (Alibaba) official model cards and technical reports",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/2407.10671v4",
      "huggingface.co/Qwen/CodeQwen1.5-7B-Chat/blob/7b0cc3380fe815e6f08fe2f80c03e05a8b1883d8/README.md",
      "huggingface.co/Qwen/QwQ-32B-Preview/blob/91906fe41a48b6a89ce2970abfd1269eefee170e/README.md",
      "huggingface.co/Qwen/QwQ-32B/blob/976055f8c83f394f35dbd3ab09a285a984907bd0/README.md",
      "huggingface.co/Qwen/Qwen1.5-110B-Chat/blob/85f86cec25901f2dbd870a86e06756903c9a876a/README.md",
      "huggingface.co/Qwen/Qwen1.5-14B-Chat/blob/9492b22871f43e975435455f5c616c77fe7a50ec/README.md",
      "huggingface.co/Qwen/Qwen1.5-32B-Chat/blob/0997b012af6ddd5465d40465a8415535b2f06cfc/README.md",
      "huggingface.co/Qwen/Qwen1.5-4B-Chat/blob/a7a4d4945d28bac955554c9abd2f74a71ebbf22f/README.md",
      "huggingface.co/Qwen/Qwen1.5-72B-Chat/blob/a9435b56d61aa68714fd7b83d6cc3e698a42c8ba/README.md",
      "huggingface.co/Qwen/Qwen1.5-7B-Chat/blob/5f4f5e69ac7f1d508f8369e977de208b4803444b/README.md",
      "huggingface.co/Qwen/Qwen2-72B-Instruct/blob/c867f763ef53f2ea9d9b31ee8501273dedd391eb/README.md",
      "huggingface.co/Qwen/Qwen2-7B-Instruct/blob/f2826a00ceef68f0f2b946d945ecc0477ce4450c/README.md",
      "huggingface.co/Qwen/Qwen2.5-0.5B-Instruct/blob/7ae557604adf67be50417f59c2c2f167def9a775/README.md",
      "huggingface.co/Qwen/Qwen2.5-1.5B-Instruct/blob/989aa7980e4cf806f80c7fef2b1adb7bc71aa306/README.md",
      "huggingface.co/Qwen/Qwen2.5-3B-Instruct/blob/aa8e72537993ba99e69dfaafa59ed015b17504d1/README.md",
      "huggingface.co/Qwen/Qwen2.5-Coder-14B-Instruct/blob/aedcc2d42b622764e023cf882b6652e646b95671/README.md",
      "huggingface.co/Qwen/Qwen2.5-VL-32B-Instruct/blob/7cfb30d71a1f4f49a57592323337a4a4727301da/README.md",
      "huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct/blob/89c86200743eec961a297729e7990e8f2ddbc4c5/README.md",
      "huggingface.co/Qwen/Qwen3-30B-A3B/blob/ad44e777bcd18fa416d9da3bd8f70d33ebb85d39/README.md",
      "huggingface.co/Qwen/Qwen3-4B/blob/1cfa9a7208912126459214e8b04321603b3df60c/README.md",
      "huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct/blob/710c13861be6c466e66de3f484069440b8f31389/README.md",
      "huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking/blob/6664affde68449468deb7527186455c7450c13c0/README.md",
      "huggingface.co/Qwen/Qwen3-VL-32B-Thinking/blob/7edd10ffd1196091948fb245ff63e406ccb2d4d1/README.md",
      "huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct/blob/2e1fd397ee46e1388853d2af2c993145b0f1098a/README.md",
      "huggingface.co/Qwen/Qwen3-VL-32B-Instruct/blob/0cfaf48183f594c314753d30a4c4974bc75f3ccb/README.md",
      "huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct/blob/9c4b90e1e4ba969fd3b5378b57d966d725f1b86c/README.md",
      "huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking/blob/d0ed0380729be07a546fdefafbb4fe411f341e92/README.md",
      "huggingface.co/Qwen/Qwen3-VL-8B-Instruct/blob/0c351dd01ed87e9c1b53cbc748cba10e6187ff3b/README.md",
      "huggingface.co/Qwen/Qwen3-VL-8B-Thinking/blob/92f3c4b4feadd3a016ef468d103bb5f58b2a2c6b/README.md",
      "huggingface.co/Qwen/Qwen3-VL-4B-Instruct/blob/ebb281ec70b05090aa6165b016eac8ec08e71b17/README.md",
      "huggingface.co/Qwen/Qwen3-VL-4B-Thinking/blob/1de27d8c51f12e819435303b9e84c4e25ba8401e/README.md",
      "huggingface.co/Qwen/Qwen3-VL-2B-Instruct/blob/89644892e4d85e24eaac8bacfd4f463576704203/README.md",
      "huggingface.co/Qwen/Qwen3-VL-2B-Thinking/blob/33e0ad94c327808ebc7d3bbc97e78e600f1eeecd/README.md",
      "huggingface.co/Qwen/CodeQwen1.5-7B/blob/5ce5a1554e50a9e3bb236de7c0b8a2a1746186e4/config.json",
      "huggingface.co/Qwen/Qwen-7B-Chat/blob/93a65d34827a3cc269b727e67004743b723e2f83/config.json",
      "huggingface.co/Qwen/Qwen-7B/blob/ef3c5c9c57b252f3149c1408daf4d649ec8b6c85/config.json",
      "huggingface.co/Qwen/Qwen2-57B-A14B-Instruct/blob/50896d66b39f1425d63720541a66c7df13e053c0/config.json",
      "huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct/blob/ea3f2471cf1b1f0db85067f1ef93848e38e88c25/README.md",
      "huggingface.co/Qwen/Qwen2.5-Coder-3B-Instruct/blob/488639f1ff808d1d3d0ba301aef8c11461451ec5/README.md",
      "huggingface.co/Qwen/Qwen2-VL-2B-Instruct/blob/895c3a49bc3fa70a340399125c650a463535e71c/config.json",
      "huggingface.co/Qwen/Qwen1.5-0.5B-Chat/blob/4d14e384a4b037942bb3f3016665157c8bcb70ea/README.md",
      "huggingface.co/Qwen/Qwen1.5-1.8B-Chat/blob/e482ee3f73c375a627a16fdf66fd0c8279743ca6/README.md",
      "huggingface.co/Qwen/Qwen1.5-MoE-A2.7B-Chat/blob/ec052fda178e241c7c443468d2fa1db6618996be/config.json",
      "huggingface.co/Qwen/Qwen2-Math-1.5B-Instruct/blob/19bd485b6474f7263545b4d5a1c29d78bcbf3a4f/config.json",
      "huggingface.co/Qwen/Qwen2-Math-7B-Instruct/blob/26843d37d3b00ee4a14263b161f312c2738660f9/config.json",
      "huggingface.co/Qwen/Qwen2-Math-72B-Instruct/blob/5c267882f3377bcfc35882f8609098a894eeeaa8/config.json",
      "huggingface.co/Qwen/Qwen2.5-Math-1.5B-Instruct/blob/aafeb0fc6f22cbf0eaeed126eff8be45b0360a35/config.json",
      "huggingface.co/Qwen/Qwen2.5-Math-7B-Instruct/blob/ef9926d75ab1d54532f6a30dd5e760355eb9aa4d/config.json",
      "huggingface.co/Qwen/Qwen2.5-Math-72B-Instruct/blob/8fcf92b1eac8eca48c3d49778ad3dd1b394c9a17/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2407.10671v4",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Qwen (Alibaba) official model cards and technical reports, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-streamlake",
    "name": "StreamLake (Kuaishou) official model documentation",
    "kind": "first-party",
    "match": [
      "streamlake.ai/document/DOC/mh1gbfvrdn6hpbzxixv"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://www.streamlake.ai/document/DOC/mh1gbfvrdn6hpbzxixv",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "StreamLake (Kuaishou) official model documentation, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "first-party-tencent-cloud",
    "name": "Tencent official model cards and Tencent Cloud documentation",
    "kind": "first-party",
    "match": [
      "cloud.tencent.com/document/product/1823/130051",
      "huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/README.md",
      "tencentcloud.com/document/product/1300/78937"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://cloud.tencent.com/document/product/1823/130051",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Tencent official model cards and Tencent Cloud documentation, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-07"
  },
  {
    "id": "first-party-tinyllama",
    "name": "TinyLlama official technical report",
    "kind": "first-party",
    "match": [
      "arxiv.org/abs/2401.02385v2"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2401.02385v2",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "TinyLlama official technical report, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "first-party-xiaomi",
    "name": "Xiaomi MiMo official model cards and documentation",
    "kind": "first-party",
    "match": [
      "huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/blob/73875d00b30a89ef8cc353a0b60b0e9f9561952d/README.md",
      "mimo.mi.com/docs/en-US/price/pay-as-you-go",
      "huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/5711b268169967567844e1e560e8a3966da959b1/README.md",
      "mimo.mi.com/docs/en-US/updates/model"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/blob/73875d00b30a89ef8cc353a0b60b0e9f9561952d/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Xiaomi MiMo official model cards and documentation, observed 2026-10-06",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "euroeval-leaderboards",
    "name": "EuroEval leaderboards (results collection), English generative leaderboard",
    "kind": "leaderboard",
    "match": [
      "github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/leaderboards/english_all.csv"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE",
    "terms_url": "https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/README.md",
    "quote": "Root LICENSE at the pinned commit: 'MIT License / Copyright (c) 2024-2026 Dan Saattrup Smart / Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software ...'. README at the same commit: 'Collection of all evaluation results from the EuroEval framework.' The repository's purpose is the results collection; no nested licence file exists under leaderboards/ (git tree checked). GitHub API licence: MIT.",
    "attribution": "EuroEval (Dan Saattrup Smart and EuroEval contributors), github.com/EuroEval/leaderboards @1a77f1d, MIT licence (copyright notice and permission notice retained). Selected cells of leaderboards/english_all.csv (Life in the UK and HellaSwag, secondary metric = accuracy); percentages divided by 100; catalogue ids mapped; split (test/val) kept as separate benchmark ids; zero-shot API runs not imported. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "scope": "Only leaderboards/english_all.csv at commit 1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0 (committed 2026-04-20; file says 'Last updated: 2026-04-17'). Git blob sha 5006507ef719ccc450afda5f498191959a129962, sha256 e9f57c98e80370153aeaed684a748fb3d7dc6c2d1672f5db6653c95117accb3e. Newer EuroEval leaderboards are generated from a storage bucket in EuroEval/EuroEval and the unlicensed HF Space EuroEval/euroeval_leaderboard; those are not covered."
  },
  {
    "id": "speech-stt-pipecat-bsd2",
    "name": "Pipecat STT Benchmark · semantic WER results table (pipecat-ai/stt-benchmark README)",
    "kind": "dataset",
    "match": [
      "github.com/pipecat-ai/stt-benchmark/blob/94f567299a21f8d617039bc69f137a131f5dedd5/README.md"
    ],
    "exact": true,
    "license": "BSD-2-Clause",
    "license_url": "https://github.com/pipecat-ai/stt-benchmark/blob/94f567299a21f8d617039bc69f137a131f5dedd5/LICENSE",
    "terms_url": "https://github.com/pipecat-ai/stt-benchmark/blob/94f567299a21f8d617039bc69f137a131f5dedd5/LICENSE",
    "quote": "Root LICENSE at the pinned commit: 'BSD 2-Clause License / Copyright (c) 2025-2026, Daily / Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice...'. Licence added by commit 27b50ee ('License under BSD-2-Clause to match the rest of pipecat-ai', 2026-08-27); repo metadata spdx_id BSD-2-Clause. README: 'The Results Summary table above is the single source of truth for the published numbers'. The table is a committed repository file; no separate data licence or carve-out found. Same licence family as the already-allowed BSD-3-Clause speech-nvidia-fastpitch-readme entry.",
    "attribution": "Pipecat STT Benchmark, Copyright (c) 2025-2026, Daily (github.com/pipecat-ai/stt-benchmark), BSD-2-Clause. Changes: selected 'Pooled WER' cells for rows mapped to exact catalogue models, percent divided by 100; retain the BSD-2-Clause notice and disclaimer. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "scope": "Only README Results Summary rows scored by the current Claude Sonnet 5.5 judge (rows marked † excluded). Ground truth is Gemini-generated and human-reviewed per README; the 1000 audio samples (pipecat-ai/smart-turn-data-v3.1-train) are not downloaded or redistributed."
  },
  {
    "id": "speech-tts-deepfake-age-ccby",
    "name": "Audio Deepfake Detection in the Age of Advanced Text-to-Speech models (arXiv:2601.20510v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2601.20510v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2601.20510v1",
    "quote": "arXiv abs page for 2601.20510v1 links http://creativecommons.org/licenses/by/4.0/; HTML header reads 'License: CC BY 4.0 arXiv:2601.20510v1 [cs.SD] 28 Jan 2026'.",
    "attribution": "Robin Singh, Aditya Yogesh Nair, Fabio Palumbo, Florian Barbaro, Anna Dyka, Lohith Rachakonda (UncovAI); 'Audio Deepfake Detection in the Age of Advanced Text-to-Speech models'; arXiv:2601.20510v1, CC BY 4.0. Changes: Table 2 WER values for the catalogued Maya1 and MeloTTS rows transcribed as printed (already fractions). No audio used. No endorsement. Added the previously uncatalogued Dia2 2B row from the same original Table 2. Exact 2B identity confirmed in section 3.1; randomized multi-speaker generation, so its SIM is intentionally absent.",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "visual-fetv-ccby",
    "name": "FETV: A Benchmark for Fine-Grained Evaluation of Open-Domain Text-to-Video Generation (arXiv:2311.01813v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2311.01813v3"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2311.01813v3",
    "quote": "arXiv abs page for 2311.01813 (current version v3) links http://creativecommons.org/licenses/by/4.0/; the v3 HTML header reads 'License: CC BY 4.0 arXiv:2311.01813v3 [cs.CV] 26 Dec 2023'. The FETV repository README (llyx97/FETV) also states 'This dataset is under CC-BY 4.0 license.'",
    "attribution": "Yuxin Liu, Lei Li, Shuhuai Ren, Rundong Gao, Shicheng Li, Sishuo Chen, Xu Sun, Lu Hou; 'FETV: A Benchmark for Fine-Grained Evaluation of Open-Domain Text-to-Video Generation', arXiv:2311.01813v3 (NeurIPS 2023 Datasets and Benchmarks), CC BY 4.0. Changes: the manual video-quality means stated in the Figure 20 caption (average of static and temporal quality, 1-5 Likert) transcribed and normalised (x-1)/4; Text2Video-zero row not imported (no catalogue identity). No endorsement.",
    "scope": "Only the authors' own manual evaluation (Sec. 4: three ratings per video from graduate students in the authors' laboratory, 619 prompts) as printed in the Figure 20 caption (A6.F20). Radar/leaderboard figures (no printed numbers), automatic metrics (FID/FVD/CLIPScore/UMT) and fine-grained alignment are not imported.",
    "verdict": "allowed",
    "checked": "2026-10-06"
  },
  {
    "id": "paper-mixtral-2401.04088v1",
    "name": "Mixtral of Experts — base-model evaluations",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2401.04088v1",
    "license_url": "/licenses/CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "match": [
      "arxiv.org/html/2401.04088v1"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "attribution": "Jiang et al., Mixtral of Experts (2024), arXiv:2401.04088v1, Table 2. CC BY 4.0. Percentages converted to fractions; derived scores are our transformation. No endorsement.",
    "quote": "Version v1 links to CC BY 4.0. Table 2 reports the authors’ own reruns of all six base models; copied leaderboard screenshots and differently configured Table 3 excluded."
  },
  {
    "id": "paper-gemma-2403.08295v4",
    "name": "Gemma report — original evaluations",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2403.08295v4",
    "license_url": "/licenses/CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "match": [
      "arxiv.org/html/2403.08295v4"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "attribution": "Gemma Team et al., Gemma: Open Models Based on Gemini Research and Technology (2024), arXiv:2403.08295v4, Table 6. CC BY 4.0. Selected original evaluations converted from percentages to fractions; no endorsement.",
    "quote": "CC BY 4.0 on exact arXiv v4. Table 6 marks author reruns with an asterisk. Only both Gemma base checkpoints and the starred Mistral 7B cells are used; copied baselines excluded."
  },
  {
    "id": "paper-tulu2-2311.10702v2",
    "name": "Tulu 2 — base and instruction-tuned evaluations",
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://arxiv.org/abs/2311.10702v2",
    "license_url": "/licenses/CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "match": [
      "arxiv.org/html/2311.10702v2"
    ],
    "exact": true,
    "icon": "arxiv.org",
    "attribution": "Ivison et al., Camels in a Changing Climate: Enhancing LM Adaptation with Tulu 2 (2023), arXiv:2311.10702v2. CC BY 4.0. Selected core capability results from Tables 1, 5 and 6; percentages converted to fractions and repeated cells deduplicated. No endorsement.",
    "quote": "Exact arXiv v2 is CC BY 4.0. Section 3.1 describes the authors evaluation framework, including dated GPT snapshots; Tables 1, 5 and 6 compare their own base and instruction-tuned runs. Imported MMLU, GSM8K, BBH, TyDiQA GP and CodexEval only. QLoRA, safety metrics and external leaderboards excluded."
  },
  {
    "id": "first-party-anthropic-models",
    "name": "Anthropic official model documentation",
    "kind": "first-party",
    "license": "Factual list price and context specification (not licensed content)",
    "terms_url": "https://platform.claude.com/docs/en/models/sonnet-5-5/overview",
    "match": [
      "platform.claude.com/docs/en/models/sonnet-5-5/overview",
      "platform.claude.com/docs/en/models/haiku-5-5/overview",
      "platform.claude.com/docs/en/about-claude/pricing"
    ],
    "exact": true,
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "anthropic.com",
    "attribution": "Anthropic official model documentation, observed 2026-10-06.",
    "quote": "Owner policy 2026-09-25: manually record the vendor’s own price and context limits as facts with observation date; benchmark numbers are excluded."
  },
  {
    "id": "stablelm-original-eval-json",
    "name": "Stability AI — original StableLM zero-shot evaluation outputs",
    "match": [
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/stablelm-base-alpha-3b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/stablelm-base-alpha-7b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/stablelm-3b-4e1t.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/EleutherAI_gpt-j-6B.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/EleutherAI_gpt-neox-20B.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/EleutherAI_pythia-12b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/EleutherAI_pythia-6.9b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/bigscience-bloom-3b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/bigscience-bloom-7b1.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/facebook-opt-2.7b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/facebook-opt-6.7b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/meta-llama-Llama-2-13b-hf.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/meta-llama-Llama-2-7b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/microsoft-phi-1_5.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/tiiuae_falcon-7b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/togethercomputer-RedPajama-INCITE-7B-Base2.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/EleutherAI-pythia-2.8b-deduped.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/Qwen-Qwen-7B-Chat.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/Qwen-Qwen-7B.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/baichuan-inc_Baichuan2-7B-Base.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/openlm-research-open_llama_3b_v2.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/openlm-research-open_llama_7b_v2.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/stablelm-base-alpha-3b-v2.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/stablelm-base-alpha-7b-v2.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/mosaicml-mpt-7b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/huggyllama-llama-7b.json",
      "github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/evals/external/kittn_mistral-7B-v0.1-hf.json"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Stability-AI/StableLM/blob/93eea082c49c61e64da1a0e1092fe82cc74b46d3/LICENSE",
    "license_url": "/licenses/stablelm-original-eval-json-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "quote": "Repository root Apache License 2.0 grants reproduction, derivative works and distribution. Scope is the committed evals/*.json output data, not the separately licensed model weights or a hosted leaderboard. README states these evaluations were performed with lm-evaluation-harness and links these full run outputs.",
    "attribution": "Stability AI and the StableLM contributors. Original zero-shot evaluation JSONs, revision 93eea082c49c61e64da1a0e1092fe82cc74b46d3. Selected accuracy/mc2 fields retained as fractions; exact model config and task version recorded. No endorsement.",
    "icon": "github.com"
  },
  {
    "id": "copen-emnlp2022-zero-shot",
    "name": "COPEN — EMNLP 2022 original zero-shot evaluations",
    "match": [
      "aclanthology.org/2022.emnlp-main.335.pdf"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "https://aclanthology.org/faq/#what-is-the-copyright-for-materials-in-the-acl-anthology",
    "license_url": "/licenses/copen-emnlp2022-zero-shot-CC-BY-4.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "quote": "ACL publication policy: Materials published in or after 2016 are licensed on a Creative Commons Attribution 4.0 International License. This is the ACL-owned 2022 paper; Tables 7 and 11 identify original evaluated checkpoints and original zero-shot measurements.",
    "attribution": "Hao Peng et al. COPEN: Probing Conceptual Knowledge in Pre-trained Language Models. EMNLP 2022, pp. 5015–5035, doi:10.18653/v1/2022.emnlp-main.335. CC BY 4.0. Selected Table 11 full-prompt (All) zero-shot measurements divided by 100; no probe-trained/fine-tuned rows or best-prompt selection. No endorsement.",
    "icon": "arxiv.org"
  },
  {
    "id": "squeezebits-exaone45-bf16",
    "name": "SqueezeBits — original EXAONE 4.5 BF16 evaluations",
    "match": [
      "github.com/SqueezeBits/FlatQuant-EXAONE4.5/blob/5f1ea65605d1f9301cd7d243b1e8a2c6590d2b46/README.md"
    ],
    "exact": true,
    "kind": "dataset",
    "license": "MIT",
    "terms_url": "https://github.com/SqueezeBits/FlatQuant-EXAONE4.5/blob/5f1ea65605d1f9301cd7d243b1e8a2c6590d2b46/LICENSE",
    "license_url": "/licenses/squeezebits-exaone45-bf16-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "quote": "MIT permission to use, copy, modify, merge, publish, distribute, sublicense and/or sell copies of the Software and associated documentation. README is committed documentation with original evaluation commands and results. Only SqueezeBits own BF16 runs are selected; no LG report values, no weight redistribution.",
    "attribution": "SqueezeBits and FlatQuant contributors, revision 5f1ea65605d1f9301cd7d243b1e8a2c6590d2b46. Selected original full-precision quality measurements; percentages divided by 100. Quantized variants are not counted as independent models to force calibration. No endorsement.",
    "icon": "github.com"
  },
  {
    "id": "magpie-v2607-card-evaluations",
    "name": "NVIDIA MagpieTTS v2607 — original multilingual evaluation",
    "match": [
      "huggingface.co/nvidia/magpie_tts_multilingual_357m/blob/19806879b16d3f2ccf28fb112b1bcd16a3c7923e/README.md"
    ],
    "exact": true,
    "kind": "first-party",
    "license": "NVIDIA Open Model License",
    "license_url": "/licenses/magpie-v2607-card-evaluations-NVIDIA.txt",
    "terms_url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "icon": "huggingface.co",
    "quote": "Section 1.3 includes documentation in Model; section 2.2 grants public display, reproduction, use, derivative works, sale and distribution. Section 3.1 requires the full agreement and NVIDIA attribution. Card explicitly declares this agreement. Only own-model numerical findings from the v2607 column; no separate dataset/audio material or v2602 checkpoint values.",
    "attribution": "NVIDIA Corporation. MagpieTTS Multilingual v2607 model card at 19806879b16d3f2ccf28fb112b1bcd16a3c7923e. Licensed by NVIDIA Corporation under the NVIDIA Open Model License. CER percentages converted to fractions; pred-context speaker similarity retained. One original multilingual study, not 26 independent studies. No endorsement."
  },
  {
    "id": "redpajama-2024-original-zero-shot",
    "name": "RedPajama — original zero-shot 3B comparison",
    "match": [
      "arxiv.org/html/2411.12372v1"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC BY 4.0",
    "license_url": "/licenses/redpajama-2024-original-zero-shot-CC-BY-4.0.txt",
    "terms_url": "https://arxiv.org/abs/2411.12372v1",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "icon": "arxiv.org",
    "quote": "The exact arXiv version explicitly declares CC BY 4.0. Selected original numerical evaluations only; no benchmark input material or external copied scores.",
    "attribution": "Maurice Weber, Daniel Y. Fu, Quentin Anthony et al. RedPajama: an Open Dataset for Training Large Language Models. arXiv:2411.12372v1, 2024. CC BY 4.0. Selected full-prompt lm-evaluation-harness columns retained as fractions, no average or HELM duplicate. No endorsement."
  },
  {
    "id": "kandinsky3-original-human-comparison",
    "name": "Kandinsky 3.0 — original human comparisons",
    "match": [
      "arxiv.org/html/2312.03511v2"
    ],
    "exact": true,
    "kind": "paper",
    "license": "CC BY 4.0",
    "license_url": "/licenses/kandinsky3-original-human-comparison-CC-BY-4.0.txt",
    "terms_url": "https://arxiv.org/abs/2312.03511v2",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "icon": "arxiv.org",
    "quote": "The exact arXiv version explicitly declares CC BY 4.0. Selected original numerical evaluations only; no benchmark input material or external copied scores.",
    "attribution": "Vladimir Arkhipkin, Andrei Filatov, Viacheslav Vasilev, Anastasia Maltseva, Said Azizov, Igor Pavlov, Julia Agafonova, Andrey Kuznetsov and Denis Dimitrov. Kandinsky 3.0 Technical Report. arXiv:2312.03511v2 (2023), CC BY 4.0. Figures 10, 13 and 16: printed outright-win fractions transcribed unchanged; tied/both/neither votes remain excluded from wins. No chart digitization or fabricated reference-model self-comparison. No endorsement."
  },
  {
    "id": "kandinsky21-card-original-fid",
    "name": "Kandinsky 2.1 card — original COCO-30k FID",
    "match": [
      "huggingface.co/kandinsky-community/kandinsky-2-1/blob/04b366659663bf60deacb96fd3e38f1cb39a3be4/README.md"
    ],
    "exact": true,
    "kind": "first-party",
    "license": "Apache-2.0",
    "license_url": "/licenses/kandinsky21-card-original-fid-Apache-2.0.txt",
    "terms_url": "https://huggingface.co/kandinsky-community/kandinsky-2-1/blob/04b366659663bf60deacb96fd3e38f1cb39a3be4/README.md",
    "verdict": "allowed",
    "checked": "2026-10-06",
    "icon": "huggingface.co",
    "quote": "Pinned README explicitly declares license: apache-2.0 and says We quantitatively measure the performance of Kandinsky 2.1 on the COCO_30k dataset, in zero-shot mode. Only its own 8.21 FID retained; competitor and older-version comparison cells not presumed to be new measurements.",
    "attribution": "Kandinsky-community / AI Forever. Original Kandinsky 2.1 zero-shot COCO-30k FID, Apache-2.0 model-card documentation. Value retained as 8.21 FID; lower is better, no percentage conversion. No endorsement."
  },
  {
    "id": "terminal-bench-leaderboard-submissions",
    "name": "Terminal-Bench 4.0 official leaderboard submissions",
    "kind": "leaderboard",
    "match": [
      "github.com/harbor-framework/terminal-bench/blob/8d0eb80ae63971cdcc89b37a672a3bfe2a6184a1/leaderboard/submissions"
    ],
    "license": "Apache-2.0",
    "terms_url": "https://github.com/harbor-framework/terminal-bench/blob/8d0eb80ae63971cdcc89b37a672a3bfe2a6184a1/LICENSE",
    "license_url": "/licenses/terminal-bench-leaderboard-submissions-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "Terminal-Bench maintainers (harbor-framework/terminal-bench), official Terminal-Bench 4.0 leaderboard submission files, Apache-2.0. Selected published accuracy per submission; model IDs mapped; percentages converted to fractions. No endorsement.",
    "quote": "Root LICENSE at the pinned commit is the Apache License, Version 2.0; the leaderboard/submissions files are in the same repository with no nested licence override.",
    "icon": "github.com"
  },
  {
    "id": "zapier-automationbench-readme",
    "name": "Zapier AutomationBench README: public-set pass rates",
    "kind": "leaderboard",
    "match": [
      "github.com/zapier/AutomationBench/blob/4a8e1061254004d9dac807054eed33fad7d1ff14/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "terms_url": "https://github.com/zapier/AutomationBench/blob/4a8e1061254004d9dac807054eed33fad7d1ff14/LICENSE",
    "license_url": "/licenses/zapier-automationbench-readme-MIT.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "Zapier, Inc., AutomationBench (github.com/zapier/AutomationBench), MIT. Selected pass rates from the README's public 600-task table; model IDs mapped; percentages converted to fractions. No endorsement.",
    "quote": "LICENSE at the pinned commit: MIT License, Copyright (c) 2026 Zapier, Inc. Its scope note: 'All benchmark code, test frameworks, simulation logic, mock data generators, configuration files, documentation, and other materials authored by Zapier are original works of authorship and are licensed to you under the MIT License.'",
    "icon": "github.com"
  },
  {
    "id": "first-party-anthropic-scores",
    "name": "Anthropic system cards and model announcements (lab-reported scores)",
    "match": [
      "www-cdn.anthropic.com/fc1b44717c85dc068bc6ba5024219938094694bd/Claude%20Opus%205.5%20System%20Card.pdf",
      "www-cdn.anthropic.com/0339e6a7c5c7b87f5c07798616dc32c215d14235/Claude%20Fable%205.1%20&%20Claude%20Mythos%205.1%20System%20Card.pdf",
      "www-cdn.anthropic.com/870c8f525702625d2c62fc6dd04c857e3250bec1/Claude%20Sonnet%205.5%20System%20Card.pdf",
      "www-cdn.anthropic.com/ceaf5c7ff2783855203fde8208ec311252dced5b/Claude%20Opus%205%20System%20Card.pdf",
      "www-cdn.anthropic.com/283ef97c476cf442c91d9a37d5b214242a55bb92/Claude%20Sonnet%205%20System%20Card.pdf",
      "www-cdn.anthropic.com/57a52ea7d8f0e54e8a542e908266086df425cdf5/Claude%20Fable%205%20&%20Claude%20Mythos%205%20System%20Card.pdf",
      "anthropic.com/claude-sonnet-5-5",
      "www-cdn.anthropic.com/0f0c97ad20d8005706296bd92aa1c27c6b2f4f61/Claude%20Opus%204.8%20System%20Card.pdf",
      "www-cdn.anthropic.com/037f06850df7fbe871e206dad004c3db5fd50340/Claude%20Opus%204.7%20System%20Card.pdf",
      "www-cdn.anthropic.com/6a5fa276ac68b9aeb0c8b6af5fa36326e0e166dd/Claude%20Opus%204.6%20System%20Card.pdf",
      "www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47/Claude%20Opus%204.5%20System%20Card.pdf",
      "www-cdn.anthropic.com/bbd8ef16d70b7a1665f14f306ee88b53f686aa75/Claude%20Sonnet%204.6%20System%20Card.pdf",
      "www-cdn.anthropic.com/e1080d6bf5ae2018ea3c2f414064be03232f5be5/Claude%20Haiku%205.5%20System%20Card.pdf",
      "www-cdn.anthropic.com/7624816413e9b4d2e3ba620c5a5e091b98b190a5/Claude%20Mythos%20Preview%20System%20Card.pdf"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://www.anthropic.com/legal/consumer-terms",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "attribution": "Anthropic PBC, benchmark results it published for its own Claude models in system cards (PDF) and announcement pages, observed 2026-10-07. Individual figures were recorded by hand and converted to fractions, with printed settings in each row's note. Lab-reported, not independent. Not licensed or endorsed by Anthropic.",
    "quote": "Owner rule (b), 2026-10-07 (docs/research/vendor-terms-2026-10-07.md). A vendor's own published score for its own model may be recorded from its announcement page, system or model card, or docs. Each figure is recorded by hand from a single fetch. Only own-model, vendor-computed cells are used, never other labs' rows or cells credited to a third party. No tables, charts or prose are reproduced, and no automated extraction is used. Governing terms: Anthropic Consumer Terms forbid 'crawl, scrape, or otherwise harvest' and bot or script access, but contain no clause against quoting an individual figure. The pages show '© 2026 Anthropic PBC', the system cards carry no licence or copyright statement, and neither grants a licence.",
    "icon": "anthropic.com"
  },
  {
    "id": "first-party-google-scores",
    "name": "Google DeepMind Gemini model page and model cards (lab-reported scores)",
    "match": [
      "deepmind.google/models/gemini",
      "deepmind.google/models/model-cards/gemini-3-8-flash",
      "deepmind.google/models/model-cards/gemini-3-7-flash",
      "deepmind.google/models/model-cards/gemini-3-5-flash",
      "deepmind.google/models/model-cards/gemini-3-5-flash-lite",
      "deepmind.google/models/model-cards/gemini-3-1-flash-lite",
      "storage.googleapis.com/deepmind-media/gemini/gemini_3_pro_model_evaluation.pdf",
      "deepmind.google/models/model-cards/gemini-3-1-pro"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://policies.google.com/terms",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": false,
    "attribution": "Google DeepMind, benchmark results it published for its own Gemini models on deepmind.google (Gemini model page, Gemini 3.8 Flash and 3.7 Flash model cards), observed 2026-10-07. Provenance and settings were checked against Google's evaluation-methodology PDFs. Individual figures were recorded by hand and converted to fractions. Lab-reported, not independent. Not licensed or endorsed by Google.",
    "quote": "Owner rule (b), 2026-10-07 (docs/research/vendor-terms-2026-10-07.md). These are Google's own scores for its own models, recorded by hand one figure at a time from a single fetch. Only Google-computed cells are used: cells the methodology PDFs credit to Vals AI, Zapier, Artificial Analysis, leaderboards or other labs are excluded. Google Terms of Service: 'You may use Google's content as allowed by these terms ... but we retain any intellectual property rights that we have in our content.' Automated access is restricted only through robots.txt, and deepmind.google has 'Allow: /'. There is no anti-quoting clause. The page footers link only Privacy and Terms, with no licence. 'exact' is false only because lib/sources.ts sourceIn() never matches an exact entry against a URL whose path ends in '/' (these canonical URLs do). The paths are page-specific.",
    "icon": "deepmind.google"
  },
  {
    "id": "google-cloud-gemini-guides",
    "name": "Google Cloud developer's guides to Gemini 3.8 Flash and 3.7 Flash",
    "match": [
      "docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-3-8-flash",
      "docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-3-7-flash"
    ],
    "kind": "first-party",
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://developers.google.com/site-policies",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "attribution": "Google Cloud, 'Developer's guide to Gemini 3.8 Flash' and 'Developer's guide to Gemini 3.7 Flash' (docs.cloud.google.com, last updated 2026-10-06), CC BY 4.0, https://creativecommons.org/licenses/by/4.0/. Changes: selected cells from the 'What's new' text tables were transcribed into JSON, percentages divided by 100 and benchmark labels mapped to our ids. Lab-reported. No endorsement.",
    "quote": "Page footer (both guides, rule a): 'Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License. For details, see the Google Developers Site Policies.' The benchmark tables are text, not images, so the grant covers them.",
    "icon": "ai.google.dev"
  },
  {
    "id": "first-party-openai-scores",
    "name": "OpenAI vendor-reported benchmark scores for its own models (openai.com announcement posts and Deployment Safety Hub system cards)",
    "match": [
      "openai.com/index/gpt-6-astra",
      "openai.com/index/introducing-gpt-6-sol-and-luna",
      "openai.com/index/introducing-gpt-6-1-sol",
      "openai.com/index/gpt-5-6",
      "openai.com/index/introducing-gpt-5-5",
      "openai.com/index/introducing-gpt-5-4",
      "openai.com/index/introducing-gpt-5-4-mini-and-nano",
      "openai.com/index/introducing-gpt-5-3-codex",
      "deploymentsafety.openai.com/gpt-6-astra",
      "deploymentsafety.openai.com/gpt-6-1-sol",
      "deploymentsafety.openai.com/gpt-5-6",
      "openai.com/index/introducing-gpt-5",
      "help.openai.com/en/articles/9624314-model-release-notes",
      "openai.com/index/gpt-4-1",
      "openai.com/index/introducing-gpt-5-2",
      "openai.com/index/introducing-gpt-5-for-developers",
      "openai.com/index/gpt-5-5-instant",
      "openai.com/index/gpt-5-1-for-developers"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://openai.com/policies/terms-of-use/",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "exact": true,
    "icon": "openai.com",
    "attribution": "OpenAI, vendor-reported results for OpenAI's own models from openai.com announcement posts and Deployment Safety Hub system cards; individual figures hand-recorded 2026-10-07 (openai.com posts read from Internet Archive captures because openai.com returns HTTP 403 to direct fetches); percentages converted to fractions. Lab-reported, not independent measurements. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07, extending the 2026-09-25 first-party-facts exception: a vendor's own published score for its own model, from its announcement, system/model card or docs, may be recorded by hand one figure at a time with the exact URL, date and printed settings; no crawling, scripts or programmatic extraction of vendor pages; no tables, charts or prose reproduced; other labs' cells and cells credited to a third party excluded; rows labelled by: 'lab'; removed if the vendor objects. OpenAI Terms of Use (Effective January 1, 2026), as quoted in docs/research/vendor-terms-2026-10-07.md, forbid users to 'Automatically or programmatically extract data or Output' and to 'Modify, copy, lease, sell or distribute any of our Services'; no clause forbids quoting an individual figure with attribution. The source URLs omit openai.com's trailing slash so that exact matching works."
  },
  {
    "id": "paper-2601.03267v2",
    "name": "OpenAI — GPT-5 System Card (arXiv:2601.03267v2)",
    "match": [
      "arxiv.org/html/2601.03267v2"
    ],
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "http://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "arxiv.org",
    "attribution": "OpenAI (2025). GPT-5 System Card. arXiv:2601.03267v2. https://arxiv.org/abs/2601.03267v2\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells and in-text figures transcribed into JSON; percentages divided by 100; MMMLU value derived as the unweighted mean of the 14 printed per-language accuracies; model names mapped to catalogue IDs (gpt-5-thinking -> gpt-5, gpt-5-thinking-mini -> gpt-5-mini, gpt-5-thinking-nano -> gpt-5-nano). No endorsement.",
    "quote": "The arXiv abstract page https://arxiv.org/abs/2601.03267 (submission history v1 19 Dec 2025, v2 1 May 2026; latest = v2) shows 'view license' linking http://creativecommons.org/licenses/by/4.0/, and the v2 HTML header reads 'License: CC BY 4.0'. Checked 2026-10-07. Rule (a): openly licensed copy, usable in full including comparison-model cells."
  },
  {
    "id": "paper-2412.16720v2",
    "name": "OpenAI — OpenAI o1 System Card (arXiv:2412.16720v2)",
    "match": [
      "arxiv.org/html/2412.16720v2"
    ],
    "kind": "paper",
    "license": "CC BY 4.0",
    "terms_url": "http://creativecommons.org/licenses/by/4.0/",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "arxiv.org",
    "attribution": "OpenAI (2024). OpenAI o1 System Card. arXiv:2412.16720v2. https://arxiv.org/abs/2412.16720v2\nCC BY 4.0: https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells and in-text figures transcribed into JSON; percentages divided by 100; MMMLU values derived as the unweighted mean of the 14 printed per-language accuracies (English row excluded); 'GPT-4o-mini' mapped to gpt-4o-mini-2024-07-18. No endorsement.",
    "quote": "The arXiv abstract page https://arxiv.org/abs/2412.16720 (submission history v1 21 Dec 2024, v2 30 Apr 2026; latest = v2) shows 'view license' linking http://creativecommons.org/licenses/by/4.0/, and the v2 HTML header reads 'License: CC BY 4.0'. Checked 2026-10-07. Rule (a): openly licensed copy, usable in full including comparison-model cells."
  },
  {
    "action": "new",
    "id": "speech-nemo-asr-scores-apache",
    "name": "NVIDIA NeMo docs: ASR scores tables (English Conformer / FastConformer)",
    "kind": "first-party",
    "match": [
      "github.com/NVIDIA/NeMo/blob/6f20e61e4d11ce4dafc0f037c3bad535e1ac3ea9/docs/source/asr/data/scores/en/conformer_en.csv",
      "github.com/NVIDIA/NeMo/blob/6f20e61e4d11ce4dafc0f037c3bad535e1ac3ea9/docs/source/asr/data/scores/en/fastconformer_en.csv"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/NVIDIA/NeMo/blob/6f20e61e4d11ce4dafc0f037c3bad535e1ac3ea9/LICENSE",
    "terms_url": "https://github.com/NVIDIA/NeMo/blob/6f20e61e4d11ce4dafc0f037c3bad535e1ac3ea9/LICENSE",
    "quote": "Apache License 2.0, §1: '\"Source\" form shall mean the preferred form for making modifications, including but not limited to software source code, documentation source, and configuration files.' §2 grants a perpetual, worldwide, royalty-free copyright license to reproduce, prepare Derivative Works of, publicly display and distribute the Work. The score CSVs are documentation source in the repository; no data carve-out.",
    "attribution": "NVIDIA Corporation, NeMo (Apache-2.0), docs/source/asr/data/scores/en at 6f20e61. Selected LibriSpeech test-clean/test-other WERs, converted to fractions. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://github.com/NVIDIA/NeMo/blob/6f20e61e4d11ce4dafc0f037c3bad535e1ac3ea9/docs/source/asr/scores.rst",
    "icon": "github.com"
  },
  {
    "action": "new",
    "id": "speech-omniasr-paper-bysa",
    "name": "Omnilingual ASR: Open-Source Multilingual Speech Recognition for 1600+ Languages (arXiv:2511.09690v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2511.09690v1"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "license_url": "http://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://arxiv.org/abs/2511.09690v1",
    "quote": "arXiv abstract page for 2511.09690v1 links the licence 'http://creativecommons.org/licenses/by-sa/4.0/'. ShareAlike applies to the adapted numerical material.",
    "attribution": "Omnilingual ASR team (Meta), Omnilingual ASR: Open-Source Multilingual Speech Recognition for 1600+ Languages, arXiv:2511.09690v1. CC BY-SA 4.0. Changes: selected FLEURS-81 and CV22-76 test mean CERs from Table 5, percent/100. This numerical adaptation is licensed CC BY-SA 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2511.09690v1",
    "icon": "arxiv.org"
  },
  {
    "action": "new",
    "id": "speech-distil-whisper-readme-mit",
    "name": "huggingface/distil-whisper README (Distil-Whisper headline WER table)",
    "kind": "dataset",
    "match": [
      "github.com/huggingface/distil-whisper/blob/a66f5c85be1e32c504e8ed68df7802d2b96db4c0/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://github.com/huggingface/distil-whisper/blob/a66f5c85be1e32c504e8ed68df7802d2b96db4c0/LICENSE",
    "terms_url": "https://github.com/huggingface/distil-whisper/blob/a66f5c85be1e32c504e8ed68df7802d2b96db4c0/LICENSE",
    "quote": "Repository LICENSE (MIT): 'Copyright 2023 The OpenAI Authors and The HuggingFace Inc. team. All rights reserved. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies...' The README is a documentation file in the repository; no data carve-out.",
    "attribution": "The HuggingFace Inc. team, Distil-Whisper (MIT), README.md at a66f5c8. Short-form WER column, percent/100. Retain the MIT notice. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/html/2311.00430v1",
    "icon": "github.com"
  },
  {
    "action": "new",
    "id": "first-party-alphacephei-vosk-models",
    "name": "Alpha Cephei Vosk models page (vendor-reported WER for its own Vosk models)",
    "kind": "first-party",
    "match": [
      "github.com/alphacep/vosk-space/blob/f10f63bc6d751cf5c4198a7dc88515f40f2d7020/models.md"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "license_url": null,
    "terms_url": "https://github.com/alphacep/vosk-space",
    "quote": "Owner rule (b), 2026-10-07: a vendor's own published score for its own model, from its docs, recorded by hand one figure at a time. The alphacep/vosk-space repository (source of alphacephei.com/vosk) has no LICENSE file, so no open licence; only Alpha Cephei's own vosk-model-* figures are used (not the third-party daanzu/zamia/aspire rows).",
    "attribution": "Alpha Cephei, Vosk models page (alphacephei.com/vosk/models; source alphacep/vosk-space models.md at f10f63b). Individual WER figures for Alpha Cephei's own Vosk models recorded by hand, percent/100. Lab-reported. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://alphacephei.com/vosk/models",
    "icon": "github.com"
  },
  {
    "action": "new",
    "id": "first-party-espnet-owsm-v4-report",
    "name": "OWSM v4 technical report (ESPnet / CMU WAVLab, arXiv:2506.00338v1): own OWSM-CTC scores",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2506.00338v1"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "license_url": null,
    "terms_url": "https://arxiv.org/abs/2506.00338v1",
    "quote": "Owner rule (b), 2026-10-07: a maker's own published score for its own model (tech report) may be recorded by hand. The arXiv page carries the default non-exclusive distribution licence (http://arxiv.org/licenses/nonexclusive-distrib/1.0/), not an open licence, so only the OWSM-CTC rows (the authors' own models) are used; Whisper, Canary, Parakeet and MMS rows are excluded.",
    "attribution": "Yifan Peng et al. (ESPnet / CMU WAVLab), OWSM v4, arXiv:2506.00338v1. Individual LibriSpeech WERs for the authors' own OWSM-CTC models recorded by hand, percent/100. Lab-reported. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2506.00338v1",
    "icon": "arxiv.org"
  },
  {
    "action": "new",
    "id": "speech-glm-asr-readme-apache",
    "name": "zai-org/GLM-ASR README benchmark table (GLM-ASR-Nano vs open/closed ASR)",
    "kind": "dataset",
    "match": [
      "github.com/zai-org/GLM-ASR/blob/a324aa59a30952c54e8097c85efea592b7fafd81/resources/bench.png"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/zai-org/GLM-ASR/blob/a324aa59a30952c54e8097c85efea592b7fafd81/LICENSE",
    "terms_url": "https://github.com/zai-org/GLM-ASR/blob/a324aa59a30952c54e8097c85efea592b7fafd81/LICENSE",
    "quote": "Repository LICENSE: 'Apache License Version 2.0, January 2004'. §1: 'Work' covers the work of authorship made available under the License; 'Object' form includes ... documentation; the README benchmark image resources/bench.png is a repository file with no separate data licence.",
    "attribution": "Z.ai (zai-org), GLM-ASR, Apache-2.0, README benchmark table resources/bench.png at a324aa5. Selected CER/WER cells for GLM-ASR-Nano and Whisper V3, percent/100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://github.com/zai-org/GLM-ASR/blob/a324aa59a30952c54e8097c85efea592b7fafd81/README.md",
    "icon": "github.com"
  },
  {
    "action": "new",
    "id": "speech-omnivoice-report-ccby",
    "name": "OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models (arXiv:2604.00688v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2604.00688v3"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2604.00688v3",
    "quote": "arXiv abstract page for 2604.00688v3 links the licence 'http://creativecommons.org/licenses/by/4.0/'.",
    "attribution": "k2-fsa OmniVoice team, OmniVoice, arXiv:2604.00688v3, CC BY 4.0. Selected Seed-TTS test-en/test-zh SIM-o and WER/CER cells from Table 1; percentages divided by 100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2604.00688v3",
    "icon": "arxiv.org"
  },
  {
    "id": "aider-code-editing-runs",
    "name": "Aider code editing benchmark runs (old edit leaderboard)",
    "kind": "dataset",
    "match": [
      "github.com/Aider-AI/aider/blob/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/aider/website/_data/edit_leaderboard.yml"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "terms_url": "https://github.com/Aider-AI/aider/blob/5dc9490bb35f9729ef2c95d00a19ccd30c26339c/LICENSE.txt",
    "license_url": "/licenses/aider-polyglot-runs-Apache-2.0.txt",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "attribution": "Paul Gauthier and Aider contributors. Selected published numerical results from the code editing leaderboard data file; model IDs mapped and percentages converted to fractions. No endorsement.",
    "quote": "Root LICENSE.txt in the same pinned repository (Apache License, Version 2.0) as this published result file; no nested override in its path. Same repository and commit as the existing aider-polyglot-runs entry.",
    "icon": "github.com"
  },
  {
    "id": "card-deepseek-ai-deepseek-v3-1-terminus",
    "name": "DeepSeek-V3.1-Terminus model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V3.1-Terminus/blob/19510d6dc61f79dbd925bd51ee8a9081c509a4b6/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1-Terminus/blob/19510d6dc61f79dbd925bd51ee8a9081c509a4b6/LICENSE",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1-Terminus/blob/19510d6dc61f79dbd925bd51ee8a9081c509a4b6/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: mit' and card text: 'This repository and the model weights are licensed under the MIT License.' The comparison table is DeepSeek's own evaluation of its two models.",
    "attribution": "DeepSeek-AI, DeepSeek-V3.1-Terminus model card (MIT). Selected published scores for DeepSeek-V3.1 and DeepSeek-V3.1-Terminus; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2411.15124v5",
    "name": "Tülu 3: Pushing Frontiers in Open Language Model Post-Training (Lambert et al., Ai2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2411.15124v5"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2411.15124v5",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "The arXiv abstract page for 2411.15124v5 links the licence 'http://creativecommons.org/licenses/by/4.0/'.",
    "attribution": "Nathan Lambert, Jacob Morrison, Valentina Pyatkin et al. (2024-2025). Tülu 3: Pushing Frontiers in Open Language Model Post-Training. arXiv:2411.15124v5. CC BY 4.0. Changes: selected numerical cells transcribed; percentages divided by 100. No endorsement.",
    "icon": "arxiv.org",
    "scope": "Table 5 (70B models, Ai2's Tülu 3 Eval suite): Tülu 3 70B, Tülu 3 70B DPO and Nemotron Llama 3.1 70B columns."
  },
  {
    "id": "wildbench-leaderboard-data",
    "name": "WildBench v2 leaderboard data (Ai2)",
    "kind": "leaderboard",
    "match": [
      "github.com/allenai/WildBench/blob/d6b8dcaf377d173d031980f97c16e1a82618c03d/leaderboard/data_dir/all_stat_wildbench.-1.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/allenai/WildBench/blob/d6b8dcaf377d173d031980f97c16e1a82618c03d/LICENSE",
    "terms_url": "https://github.com/allenai/WildBench/blob/d6b8dcaf377d173d031980f97c16e1a82618c03d/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Root LICENSE in the same pinned repository (Apache License, Version 2.0) as the leaderboard data file; no nested licence override in leaderboard/data_dir. GitHub reports the repository licence as apache-2.0.",
    "attribution": "Bill Yuchen Lin et al. (Ai2), WildBench. Apache-2.0. Selected published WB-Score values; model names mapped. No endorsement.",
    "icon": "github.com"
  },
  {
    "id": "card-huggingfaceh4-zephyr-orpo-141b-a35b-v0-1",
    "name": "HuggingFaceH4 zephyr-orpo-141b-A35b-v0.1 model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1/blob/a3be084543d278e61b64cd600f28157afc79ffd3/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/HuggingFaceH4/zephyr-orpo-141b-A35b-v0.1/blob/a3be084543d278e61b64cd600f28157afc79ffd3/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: apache-2.0' in the same pinned repository that holds the README; the Performance table states all rows were evaluated by H4 with LightEval.",
    "attribution": "Hugging Face H4, Zephyr 141B-A39B model card (Apache-2.0). Selected scores; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "internlm-github-model-cards",
    "name": "InternLM GitHub repository model cards (Shanghai AI Laboratory)",
    "kind": "first-party",
    "match": [
      "github.com/InternLM/InternLM/blob/68fdc71330059255fd647abf28219ad11e8e4865/model_cards/internlm2.5_7b.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/InternLM/InternLM/blob/68fdc71330059255fd647abf28219ad11e8e4865/LICENSE",
    "terms_url": "https://github.com/InternLM/InternLM/blob/68fdc71330059255fd647abf28219ad11e8e4865/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Root LICENSE of the same pinned repository is the Apache License 2.0 (GitHub reports apache-2.0); the model card markdown is a repository file with no nested licence override.",
    "attribution": "Shanghai AI Laboratory / InternLM team, InternLM GitHub model card (Apache-2.0). Selected OpenCompass scores; percentages converted to fractions. No endorsement.",
    "icon": "github.com"
  },
  {
    "id": "instruct-eval-readme",
    "name": "InstructEval (DeCLaRe Lab) README results table",
    "kind": "leaderboard",
    "match": [
      "github.com/declare-lab/instruct-eval/blob/49bbd2ce7e3a5bdc4024ba62a5c9c68c107ee966/README.md"
    ],
    "exact": true,
    "license": "MIT / Apache-2.0",
    "license_url": "https://github.com/declare-lab/instruct-eval/blob/49bbd2ce7e3a5bdc4024ba62a5c9c68c107ee966/LICENSE",
    "terms_url": "https://github.com/declare-lab/instruct-eval/blob/49bbd2ce7e3a5bdc4024ba62a5c9c68c107ee966/LICENCE",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Root of the same pinned repository holds LICENSE ('MIT License, Copyright (c) 2023 Deep Cognition and Language Research (DeCLaRe) Lab') and LICENCE (Apache License 2.0); GitHub reports apache-2.0. Both permit reuse of the README's results table.",
    "attribution": "Yew Ken Chia, Pengfei Hong, Lidong Bing, Soujanya Poria (DeCLaRe Lab), InstructEval. Selected results; percentages converted to fractions; model paths mapped. No endorsement.",
    "icon": "github.com"
  },
  {
    "id": "card2-tencent-hy4-preview",
    "name": "Tencent Hy4 preview model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/README.md#benchmark-appendix"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/LICENSE",
    "terms_url": "https://huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: apache-2.0' in the same pinned repository; the benchmark appendix image is a committed file (assets/benchmark-appendix.jpg) of that repository. Match URL carries the card's evaluation-section fragment so these score rows stay separate from the existing first-party price/context entry that matches the bare README URL.",
    "attribution": "Tencent Hy team, Hy4 preview model card (Apache-2.0). Selected scores visually read from the benchmark appendix image; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "card2-qwen-qwen2-5-vl-32b-instruct",
    "name": "Qwen2.5-VL-32B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen2.5-VL-32B-Instruct/blob/7cfb30d71a1f4f49a57592323337a4a4727301da/README.md#evaluation"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen2.5-VL-32B-Instruct/blob/7cfb30d71a1f4f49a57592323337a4a4727301da/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: apache-2.0' in the same pinned repository that holds the README. Match URL carries the card's evaluation-section fragment so these score rows stay separate from the existing first-party price/context entry that matches the bare README URL.",
    "attribution": "Qwen team (Alibaba), Qwen2.5-VL-32B-Instruct model card (Apache-2.0). Selected scores; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "glm-v-github-bench-image",
    "name": "Z.ai GLM-V GitHub repository benchmark image (GLM-4.6V)",
    "kind": "first-party",
    "match": [
      "github.com/zai-org/GLM-V/blob/9013a771cfe578d8a0c7dfc4548156f3a23e4890/resources/bench_46v.jpeg"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/zai-org/GLM-V/blob/9013a771cfe578d8a0c7dfc4548156f3a23e4890/LICENSE",
    "terms_url": "https://github.com/zai-org/GLM-V/blob/9013a771cfe578d8a0c7dfc4548156f3a23e4890/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Root LICENSE of the same pinned repository is the Apache License 2.0 (GitHub reports apache-2.0); the benchmark image is a committed repository file (resources/bench_46v.jpeg).",
    "attribution": "Z.ai (Zhipu AI) GLM-V team, GLM-4.6V benchmark table image (Apache-2.0 repository). Selected own-model scores visually read; percentages converted to fractions. No endorsement.",
    "icon": "github.com"
  },
  {
    "id": "card2-qwen-qwen3-vl-235b-a22b-instruct",
    "name": "Qwen3-VL-235B-A22B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct/blob/710c13861be6c466e66de3f484069440b8f31389/README.md#model-performance"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct/blob/710c13861be6c466e66de3f484069440b8f31389/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: apache-2.0' in the same pinned repository that holds the README with the embedded benchmark tables. Match URL carries the card's evaluation-section fragment so these score rows stay separate from the existing first-party price/context entry that matches the bare README URL.",
    "attribution": "Qwen team (Alibaba), Qwen3-VL-235B-A22B-Instruct model card (Apache-2.0). Selected own-model scores visually read from the embedded table; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "card2-qwen-qwen3-vl-235b-a22b-thinking",
    "name": "Qwen3-VL-235B-A22B-Thinking model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking/blob/6664affde68449468deb7527186455c7450c13c0/README.md#model-performance"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking/blob/6664affde68449468deb7527186455c7450c13c0/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: apache-2.0' in the same pinned repository that holds the README with the embedded benchmark tables. Match URL carries the card's evaluation-section fragment so these score rows stay separate from the existing first-party price/context entry that matches the bare README URL.",
    "attribution": "Qwen team (Alibaba), Qwen3-VL-235B-A22B-Thinking model card (Apache-2.0). Selected own-model scores visually read from the embedded table; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "granite-3-0-paper-github",
    "name": "IBM Granite 3.0 Language Models technical report (GitHub)",
    "kind": "paper",
    "match": [
      "github.com/ibm-granite/granite-3.0-language-models/blob/5960b7fb3308588bf5f86348278ec9d6015ab647/paper.pdf"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/ibm-granite/granite-3.0-language-models/blob/5960b7fb3308588bf5f86348278ec9d6015ab647/LICENSE",
    "terms_url": "https://github.com/ibm-granite/granite-3.0-language-models/blob/5960b7fb3308588bf5f86348278ec9d6015ab647/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Root LICENSE of the same pinned repository that holds paper.pdf is the Apache License 2.0 (GitHub reports Apache-2.0); no nested licence notice for the paper.",
    "attribution": "IBM Granite Team, Granite 3.0 Language Models (paper.pdf, Apache-2.0 repository). Selected Table 10 scores; percentages converted to fractions. No endorsement.",
    "icon": "github.com"
  },
  {
    "id": "first-party-lgai-scores",
    "name": "LG AI Research vendor-reported benchmark scores for its own EXAONE models (Hugging Face model cards)",
    "match": [
      "huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B/blob/a1d54d1c148c30881ed27e035b650da489b51b92/README.md#performance",
      "huggingface.co/LGAI-EXAONE/EXAONE-4.5-33B/blob/570aa4b15a4f45ba1133072b45f50198f6e3b4fd/README.md#evaluation-results",
      "huggingface.co/LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct/blob/ccce25bd39c141fe053e0bc75818a8f5fe962802/README.md#evaluation",
      "huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B/blob/61e6d578eb102b578e5704e2916ac841df9eca0a/README.md#evaluation-results",
      "huggingface.co/LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct/blob/553ea250b9a5317231459279d5847d6cf955b9aa/README.md#evaluation",
      "huggingface.co/LGAI-EXAONE/EXAONE-3.5-32B-Instruct/blob/3d27aab477c042df2e174d36eea47b77c8871ef0/README.md#evaluation",
      "huggingface.co/LGAI-EXAONE/EXAONE-Deep-32B/blob/4e9a4c20b6c3f9093777bfe6a98b2be508abb919/README.md#evaluation"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B/blob/a1d54d1c148c30881ed27e035b650da489b51b92/LICENSE",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "huggingface.co",
    "attribution": "LG AI Research, vendor-reported results for LG's own EXAONE models from its pinned Hugging Face model cards; individual figures hand-recorded 2026-10-07; percentages converted to fractions. Lab-reported, not independent measurements. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. Match URLs carry the cards' evaluation-section fragments so these score rows stay separate from the existing first-party-lg-ai-research price/context entry that matches the bare README URLs."
  },
  {
    "id": "first-party-amazon-scores",
    "name": "Amazon vendor-reported benchmark scores for its own Nova models (Amazon Science technical reports)",
    "match": [
      "cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://www.amazon.com/gp/help/customer/display.html?nodeId=508088",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "amazon.science",
    "attribution": "Amazon AGI, vendor-reported results for Amazon's own Nova models; individual figures hand-recorded 2026-10-07; percentages converted to fractions. Lab-reported, not independent measurements. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model, from its announcement, system/model card or docs, may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'; removed if the vendor objects. The Nova Premier report carries no open licence. Terms: Amazon Conditions of Use, linked from the amazon.science footer."
  },
  {
    "id": "first-party-arcee-scores",
    "name": "Arcee AI vendor-reported benchmark scores for its own Trinity models (Hugging Face model cards)",
    "match": [
      "huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/dc6a99b6b74202880e31491a445a94884c97890c/README.md#benchmarks",
      "huggingface.co/arcee-ai/Trinity-Large-Preview/blob/d8e126719c8bf2b3ac5a3551bd1f07fcee767a90/README.md#benchmarks"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/dc6a99b6b74202880e31491a445a94884c97890c/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "huggingface.co",
    "attribution": "Arcee AI, vendor-reported results for Arcee's own Trinity models from its pinned Hugging Face cards; individual figures hand-recorded 2026-10-07; percentages converted to fractions. Lab-reported. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. Cards are tagged openmdw-1.1; recorded under rule (b) rather than as licensed copies. Match URLs carry the '#benchmarks' fragment so the score rows stay separate from the existing first-party-arcee price/context entry."
  },
  {
    "id": "first-party-mistral-scores",
    "name": "Mistral AI vendor-reported benchmark scores for its own models (Hugging Face model cards)",
    "match": [
      "huggingface.co/mistralai/Ministral-8B-Instruct-2410/blob/2f494a194c5b980dfb9772cb92d26cbb671fce5a/README.md#benchmarks",
      "huggingface.co/mistralai/Mathstral-7B-v0.1/blob/ec3a48484ef241dfe03282edcb0f25e564923823/README.md#evaluation",
      "huggingface.co/mistralai/Pixtral-Large-Instruct-2411/blob/c1e51f6f11974a1199685d35c62f5a425c2d001e/README.md#metrics",
      "huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503/blob/68faf511d618ef198fef186659617cfd2eb8e33a/README.md#vision",
      "huggingface.co/mistralai/Mistral-Medium-3.5-128B/blob/22b2b868a15677cfa6061277ed2f653d1349a9ab/README.md#agentic-benchmarks"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://mistral.ai/terms/",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "huggingface.co",
    "attribution": "Mistral AI, vendor-reported results for Mistral's own models from its pinned Hugging Face cards; individual figures hand-recorded 2026-10-07; percentages converted to fractions. Lab-reported. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. Match URLs carry section fragments so the score rows stay separate from the existing first-party-mistral price/context entry."
  },
  {
    "id": "first-party-nvidia-scores",
    "name": "NVIDIA vendor-reported benchmark scores for its own Nemotron models (Hugging Face model cards)",
    "match": [
      "huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16/blob/dfaf35de3e30f1867dd8dbc38a7fc9fb52d3914f/README.md#evaluation-results",
      "huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2/blob/6533e8de2c68e4536bf7c411d7a3ce5734111476/README.md#evaluation-results",
      "arxiv.org/html/2512.20848v1#S3.T3",
      "arxiv.org/html/2604.12374v1#S3.T5",
      "huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/README.md#reasoning-benchmark-evaluations",
      "huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/blob/2dc98e2afe4face0e4ce40972a915c45368bd34a/README.md#benchmarks",
      "huggingface.co/nvidia/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16/blob/8fe5546888e9bd03fdbf52d808adebdfca901b52/README.md#benchmarks"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16/blob/dfaf35de3e30f1867dd8dbc38a7fc9fb52d3914f/README.md",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "huggingface.co",
    "attribution": "NVIDIA, vendor-reported results for NVIDIA's own Nemotron models from pinned Hugging Face cards; individual figures hand-recorded 2026-10-07; percentages converted to fractions. Lab-reported. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time; one vendor entry groups that vendor's pages into one evaluator. Match URLs carry the '#evaluation-results' fragment so these rows stay separate from the existing per-card entries (card-nvidia-nvidia-nemotron-3-nano-4b-bf16, card-nvidia-nvidia-nemotron-nano-9b-v2) that match the bare README URLs."
  },
  {
    "id": "card2-qwen-qwen2-0-5b-instruct",
    "name": "Qwen2-0.5B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen2-0.5B-Instruct/blob/c540970f9e29518b1d8f06ab8b24cba66ad77b6d/README.md#evaluation"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/Qwen/Qwen2-0.5B-Instruct/blob/c540970f9e29518b1d8f06ab8b24cba66ad77b6d/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: apache-2.0' in the same pinned repository that holds the README. Match URL carries the '#evaluation' fragment to keep these rows separate from any price/context entry on the bare README URL.",
    "attribution": "Qwen team (Alibaba), Qwen2-0.5B-Instruct model card (Apache-2.0). Selected scores; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "card2-openchat-openchat-3-5",
    "name": "OpenChat 3.5 model card (openchat/openchat_3.5)",
    "kind": "first-party",
    "match": [
      "huggingface.co/openchat/openchat_3.5/blob/0fc98e324280bc4bf5d2c30ecf7b97b84fb8a19b/README.md#benchmarks"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/openchat/openchat_3.5/blob/0fc98e324280bc4bf5d2c30ecf7b97b84fb8a19b/README.md",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "Card YAML 'license: apache-2.0' in the same pinned repository that holds the README; the Benchmarks section states all models were evaluated by the OpenChat team except rows marked as taken from reports.",
    "attribution": "OpenChat team (Guan Wang et al.), OpenChat 3.5 model card (Apache-2.0). Selected scores; percentages converted to fractions. No endorsement.",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2404.06395v3",
    "name": "MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies (Hu et al.)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2404.06395v3"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2404.06395v3",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "The arXiv abstract page for 2404.06395v3 links the licence 'http://creativecommons.org/licenses/by/4.0/'.",
    "attribution": "Shengding Hu et al. (2024). MiniCPM. arXiv:2404.06395v3. CC BY 4.0. Selected Table 3 cells transcribed; percentages divided by 100. No endorsement.",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2311.11045v2",
    "name": "Orca 2: Teaching Small Language Models How to Reason (Mitra et al., Microsoft)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2311.11045v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2311.11045v2",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "quote": "The arXiv abstract page for 2311.11045 links the licence 'http://creativecommons.org/licenses/by/4.0/'.",
    "attribution": "Arindam Mitra et al. (2023). Orca 2. arXiv:2311.11045v2. CC BY 4.0. Selected Table 1/2 cells transcribed; percentages divided by 100. No endorsement.",
    "icon": "arxiv.org"
  },
  {
    "action": "new",
    "id": "visual-rapid-i2v-realism",
    "name": "Rapidata/image-to-video-realism published dataset-card rankings",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/Rapidata/image-to-video-realism/blob/d7abf89e71cb45fff0b08d70df389026cb78024f/README.md"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "Dataset card front matter at the pinned revision: license: 'cc-by-4.0' (not gated).",
    "attribution": "Rapidata. Rapidata image-to-video realism & visual quality Benchmark, Hugging Face dataset Rapidata/image-to-video-realism @ d7abf89e71cb45fff0b08d70df389026cb78024f (2026-10-01), CC BY 4.0. Extracted the published Overall ranking Elo values; model slugs mapped only to the exact named model. No values estimated or rerun. No endorsement; provided as-is.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://huggingface.co/datasets/Rapidata/image-to-video-realism/blob/d7abf89e71cb45fff0b08d70df389026cb78024f/README.md",
    "icon": "huggingface.co",
    "scope": "Rapidata's own human pairwise study (2,981 head-to-head rows, 522,210 votes, 19 participants, 25 image prompts); only the 'Overall ranking' ELO table is imported. The Photorealism and Visual integrity rankings are sub-leaderboards of the same comparisons and are not imported separately. Withheld: ground_truth_baseline (not a model), hidream-o1-video (no Elo printed, '—'), minimax-h3-max, cosmos-3-super (video), alaya-world, yume-1-5, sana-wm (not in catalogue)."
  },
  {
    "action": "new",
    "id": "visual-rapid-relative-camera",
    "name": "Rapidata/relative-camera-movement-human-preference-900k published dataset-card rankings",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/Rapidata/relative-camera-movement-human-preference-900k/blob/e03ac787593bada12ac6b088d3d35da4716517c7/README.md"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "Dataset card front matter: license: 'cc-by-4.0'; Licensing & attribution section: \"Prompts. Written by the dataset creators, released under CC-BY-4.0.\" ... \"Human annotations. Collected via Rapidata, released under CC-BY-4.0.\"",
    "attribution": "Rapidata. Rapidata Relative Camera Movement Benchmark, Hugging Face dataset Rapidata/relative-camera-movement-human-preference-900k @ e03ac787593bada12ac6b088d3d35da4716517c7 (2026-09-29), CC BY 4.0. Extracted the published Alignment-leaderboard Elo values; model labels mapped only to the exact named model. No values estimated or rerun. No endorsement; provided as-is.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://huggingface.co/datasets/Rapidata/relative-camera-movement-human-preference-900k/blob/e03ac787593bada12ac6b088d3d35da4716517c7/README.md",
    "icon": "huggingface.co",
    "scope": "Rapidata's own human pairwise study (22,691 comparisons, 900,318 votes, 15 models, 220 hand-written prompts on 50 stills). Only the 'Overall ranking (ELO on the Alignment leaderboard)' table is imported; the win-share breakdown table is derived from the same votes. Withheld: MiniMax H3 Max, Cosmos Predict 2.5, Yume 1.5 (not in catalogue)."
  },
  {
    "action": "new",
    "id": "visual-heim",
    "name": "HEIM: Holistic Evaluation of Text-to-Image Models (arXiv 2311.04287v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2311.04287v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2311.04287v1 links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2311.04287v1 [cs.CV] 07 Nov 2023\". Datasheet Q49: \"We own the metadata and release as CC-BY-4.0.\"",
    "attribution": "Tony Lee, Michihiro Yasunaga, Chenlin Meng, Yifan Mai, Joon Sung Park, Agrim Gupta, Yunzhi Zhang, Deepak Narayanan, Hannah Benita Teufel, Marco Bellagente, Minguk Kang, Taesung Park, Jure Leskovec, Jun-Yan Zhu, Li Fei-Fei, Jiajun Wu, Stefano Ermon, Percy Liang. Holistic Evaluation of Text-to-Image Models, arXiv:2311.04287v1 (7 Nov 2023), CC BY 4.0. Extracted Table 5 Alignment win rates for catalog models; No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2311.04287v1",
    "icon": "arxiv.org",
    "scope": "HEIM authors' own evaluation of 26 models (Sec. 6-7, Table 5 'Result summary', aspect win rates). Only the Alignment column is imported (one benchmark per table; HEIM gives no overall score, and Quality/Aesthetics/Reasoning/Knowledge etc. are other aspect columns of the same summary). Withheld rows: 'Stable Diffusion v2 base' and 'v2.1 base' (512px stabilityai/stable-diffusion-2-base and 2-1-base checkpoints per Appendix D, not the 768-v sd-2-0/sd-2-1 catalog entries), 'Openjourney v2' (= Openjourney v4 repo, not in catalog), minDALL-E (Kakao Brain; not in catalog), DALL-E mini, Dreamlike Diffusion 1.0, Redshift, Vintedois, Safe SD x4, Promptist, Lexica, MultiFusion, GigaGAN (not in catalog). The crfm.stanford.edu website data are not used."
  },
  {
    "action": "new",
    "id": "visual-imagereward",
    "name": "ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation (arXiv 2304.05977v4)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2304.05977v4"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2304.05977v4 links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads \"License: CC BY 4.0 arXiv:2304.05977v4 [cs.CV] 28 Dec 2023\".",
    "attribution": "Jiazheng Xu, Xiao Liu, Yuchen Wu, Yuxuan Tong, Qinkai Li, Ming Ding, Jie Tang, Yuxiao Dong. ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generation, arXiv:2304.05977v4 (28 Dec 2023), CC BY 4.0. Extracted Table 1 human-evaluation win counts and author-computed zero-shot COCO FID for catalog models; No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2304.05977v4",
    "icon": "arxiv.org",
    "scope": "Authors' own Table 1 (Sec. 2.3) model comparison. Imported: Real-User-Prompts 'Human Eval. #Win' (researcher annotation) and MS-COCO 2014 'Zero-shot FID' computed by the authors (two separate experiments in one table). Not imported: ImageReward/CLIP score columns (automatic metrics, correlated with the human run / same images). Withheld rows: DALL-E 2 FID (caption: 'taken from [42]', copied), 'Stable Diffusion 2.1-base' (512px base checkpoint, not the 768-v sd-2-1 catalog entry), Openjourney (footnote points to the website openjourney.art, checkpoint/version (v1 vs v4) not established). CogView 2 rows marked lab (co-authors Ming Ding, Jie Tang are CogView2 authors)."
  },
  {
    "action": "new",
    "id": "visual-vvrbench",
    "name": "Verifiable Visual Rewards Transfer from Synthetic Scenes to Natural Prompts (VVRBench), arXiv:2609.35641v1",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.35641v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2609.35641v1 (v1, 28 Sep 2026, only version) links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2609.35641v1 [cs.AI] 28 Sep 2026'.",
    "attribution": "Shuyue Stella Li, Xiaochuang Han, Yulia Tsvetkov, Luke Zettlemoyer. Verifiable Visual Rewards Transfer from Synthetic Scenes to Natural Prompts, arXiv:2609.35641v1 (28 Sep 2026), CC BY 4.0. Extracted Table 3 (VVRBench) and Table 4 (VVRBench-Challenge) overall accuracies, divided by 100; model labels mapped to exact catalogue models via Table 9. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2609.35641v1",
    "icon": "arxiv.org",
    "scope": "The authors' own generations and verifier scoring. Imported: Table 3 overall accuracy (VVRBench, 10,000 tasks) and Table 4 overall accuracy (VVRBench-Challenge, 720 tasks). Not imported: complexity-bucket columns (sub-scores of the same run), Table 10 VVRBench-Fast (an 820-task subset of the main split; would double-count GPT-Image-2 / GPT-Image-1-mini), per-constraint Tables 11-12, and the authors' own RLVVR post-trained SD3.5 checkpoints (Tables 5, 14-16). Withheld row: Sana 1.6B (not in catalogue)."
  },
  {
    "action": "new",
    "id": "visual-chartsync",
    "name": "ChartSync: A Benchmark for Visuo-Logical Cascading Chart Editing (arXiv:2607.10301v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2607.10301v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2607.10301v1 (v1, 11 Jul 2026) links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2607.10301v1 [cs.CV] 11 Jul 2026'.",
    "attribution": "Jiakang Yu, Yixuan Chai, Tianci Wang, Rihui Jin, Guangkai Xu, Hongtao Deng, Xun Zhu, Wang Gao, Xinrun Guo, Haipang Wu et al. ChartSync: A Benchmark for Visuo-Logical Cascading Chart Editing, arXiv:2607.10301v1 (11 Jul 2026), CC BY 4.0. Extracted Table 2 Overall scores, divided by 100. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2607.10301v1",
    "icon": "arxiv.org",
    "scope": "Authors' own edits (open models run locally at Table 7 settings, proprietary models via API) and scoring. Only Table 2 'Overall' imported; per-metric columns and VLM Avg are components of it. Withheld: Code-Mediated Pipeline (not a model), DeepGen-1.0, Intern-VL-U, FireRed-Image-Edit-1.0/1.1 (not in catalogue), FLUX.2-Klein-Base-9B (base checkpoint, not the distilled flux-2-klein-9b catalogue entry)."
  },
  {
    "action": "new",
    "id": "visual-infoedit",
    "name": "InfoEdit: Probing Global Layout Reasoning in Infographic Editing (arXiv:2609.33286v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.33286v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2609.33286v1 (v1, 27 Sep 2026) links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2609.33286v1 [cs.CV] 27 Sep 2026'.",
    "attribution": "Cheng Yang, Chufan Shi, Huijuan Wang, Bo Shui, Yaokang Wu, Muzi Tao, Yibo Yan, Xuezhe Ma, Taylor Berg-Kirkpatrick. InfoEdit: Probing Global Layout Reasoning in Infographic Editing, arXiv:2609.33286v1 (27 Sep 2026), CC BY 4.0. Extracted Table 2 'Avg.' success rates, divided by 100; model identities from Table 17. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2609.33286v1",
    "icon": "arxiv.org",
    "scope": "Authors' own edits and judging. Only Table 2 'Avg.' (mean SR) imported; per-task EC/CP/SR are components. Code-level editing (Table 3) and hint experiments (Table 4) not imported. Withheld: FLUX.2-klein-base-9B (base checkpoint, not the distilled catalogue entry)."
  },
  {
    "action": "new",
    "id": "visual-paintbench",
    "name": "PaintBench: Deterministic Evaluation of Precise Visual Editing (arXiv:2606.00188v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.00188v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2606.00188v1 (v1, 29 May 2026) links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2606.00188v1 [cs.GR] 29 May 2026'.",
    "attribution": "Kai Xu, Ellis Brown, Shrikar Madhu, Rob Fergus, He He, Saining Xie (New York University). PaintBench: Deterministic Evaluation of Precise Visual Editing, arXiv:2606.00188v1 (29 May 2026), CC BY 4.0. Extracted Table 1 'Benchmark Avg.' mIoU, divided by 100. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2606.00188v1",
    "icon": "arxiv.org",
    "scope": "Authors' own edits and deterministic pixel scoring. Only Table 1 'Benchmark Avg.' imported (per-task and category rows are components). TinyGrafixBench (Table 3), a companion data-visualisation suite built to correlate with PaintBench, and the visual-condition study (Table 2) are not imported. Withheld: HunyuanImage-3.0 (run with 8-step distilled sampling for editing; base vs -Instruct / distilled checkpoint not established)."
  },
  {
    "action": "new",
    "id": "visual-mechverse",
    "name": "MechVerse: Evaluating Physical Motion Consistency in Video Generation Models (arXiv:2605.14843v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.14843v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2605.14843v1 (v1, 14 May 2026) links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2605.14843v1 [cs.CV] 14 May 2026'.",
    "attribution": "Rahul Jain, Mayank Patel, Asim Unmesh, Karthik Ramani (Purdue University). MechVerse: Evaluating Physical Motion Consistency in Video Generation Models, arXiv:2605.14843v1 (14 May 2026), CC BY 4.0. Extracted Table 1 WorldModelBench 'Sum' scores, divided by 10. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2605.14843v1",
    "icon": "arxiv.org",
    "scope": "Authors' own generations on the MechVerse test split, scored with the WorldModelBench judge. Only Table 1 'Sum' imported; VBench-I2V columns and the WorldModelBench sub-scores are not. Withheld: CogVideoX-1.5 (5B) (I2V checkpoint not in catalogue), LTX-2 (22B) (LTX-2 is 19B; 22B matches LTX-2.3, so identity unclear), LTX-Video (13B) (0.9.7/0.9.8 13B not in catalogue), VideoCrafter (version unstated), DynamiCrafter, ConsistI2V (not in catalogue), and the authors' MechVerse fine-tunes."
  },
  {
    "action": "new",
    "id": "visual-rapid-benchai-image",
    "name": "Rapidata/text-to-image-human-preference-6M (Benchmark.ai Image Benchmark) published dataset-card rankings",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/Rapidata/text-to-image-human-preference-6M/blob/2f602f465ad418ac886f2fde92ba42b85e1369db/README.md"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "Dataset card front matter at revision 2f602f465ad418ac886f2fde92ba42b85e1369db: 'license: cc-by-4.0'. Licensing & attribution section: 'Human annotations. Collected via Rapidata, released under CC-BY-4.0.' (Rapidata-original and community prompts also CC-BY-4.0; generated images governed by each provider's terms and not used.)",
    "attribution": "Rapidata. Benchmark.ai Image Benchmark, Hugging Face dataset Rapidata/text-to-image-human-preference-6M @ 2f602f465ad418ac886f2fde92ba42b85e1369db (2026-10-01), CC BY 4.0. Extracted the published Overall ranking ELO values; model labels mapped only to exact catalogue models. No values estimated or rerun. No endorsement; provided as-is.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://huggingface.co/datasets/Rapidata/text-to-image-human-preference-6M/blob/2f602f465ad418ac886f2fde92ba42b85e1369db/README.md",
    "icon": "huggingface.co",
    "scope": "Rapidata's own human pairwise study (43 T2I models). Only the 'Overall ranking' ELO column is imported; the Alignment/Preference/Coherence/Text coherence columns and the by-output-type/capability tables are slices of the same votes. The benchmark.ai website is not used. Withheld: Grok Image Imagine 2.0 (Medium); Qwen-Image-3.0; Grok Image Imagine 2.0 (Thinking); Imagen 3; Imagen 4 Fast; Pruna-P-Ideogram (High); Pruna-P-Ideogram (Low); Pruna-P-Ideogram (Medium); Pruna-P-Ideogram (Very Low) (setting variants without a separate catalogue entry, models not in the catalogue, or identity unclear: 'Qwen-Image-3.0' lacks the 'Pro' that the card uses for other Pro tiers, so it is not mapped to qwen-image-3-0-pro; 'Imagen 3' does not say imagen-3.0-generate-001 vs -002)."
  },
  {
    "action": "new",
    "id": "visual-rapid-face",
    "name": "Rapidata/Face_Generation_Benchmark (Human Face Generation Alignment) published dataset-card leaderboard",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/Rapidata/Face_Generation_Benchmark/blob/c6ed7eab14620db5daa77aec0feda5992091f5b3/README.md"
    ],
    "exact": true,
    "license": "CDLA-Permissive-2.0",
    "license_url": "/licenses/visual-rapid-face-CDLA-Permissive-2.0.txt",
    "terms_url": "https://cdla.dev/permissive-2-0/",
    "quote": "Dataset card front matter at revision c6ed7eab14620db5daa77aec0feda5992091f5b3: 'license: cdla-permissive-2.0'. CDLA-Permissive-2.0 s.1.1: 'A Data Recipient may use, modify, and share the Data made available by Data Provider(s) under this agreement if that Data Recipient follows the terms of this agreement.' s.2.1: 'A Data Recipient may share Data, with or without modifications, so long as the Data Recipient makes available the text of this agreement with the shared Data.' s.3.1: 'This agreement does not impose any restriction or obligations with respect to the use, modification, or sharing of Results.' Keep the agreement text in licenses/.",
    "attribution": "Rapidata. Rapidata Human Face Generation Alignment, Hugging Face dataset Rapidata/Face_Generation_Benchmark @ c6ed7eab14620db5daa77aec0feda5992091f5b3, CDLA-Permissive-2.0. Extracted the published leaderboard scores; Replicate model ids mapped only to exact catalogue models. No values estimated or rerun. No endorsement; provided as-is.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://huggingface.co/datasets/Rapidata/Face_Generation_Benchmark/blob/c6ed7eab14620db5daa77aec0feda5992091f5b3/README.md",
    "icon": "huggingface.co",
    "scope": "Rapidata's own human pairwise study (12 T2I models). Only the card's leaderboard 'score' column imported. The app.rapidata.ai live leaderboard is not used. Withheld (not in catalogue): minimax/image-01; google/imagen-4-fast; prunaai/wan-2.2-image; bria/image-3.2; luma/photon-flash."
  },
  {
    "action": "new",
    "id": "visual-videoargus",
    "name": "VideoArgus: Agentic Rubric-Grounded Unified Evaluation for Video Generation and Editing (arXiv:2608.05485v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2608.05485v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2608.05485v1 (v1, 6 Aug 2026) links the licence http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2608.05485v1 [cs.CV] 06 Aug 2026'.",
    "attribution": "Ziyun Zeng, Zixuan Wang, Yongsheng Yu, Hang Hua, Jiebo Luo (University of Rochester). VideoArgus: Agentic Rubric-Grounded Unified Evaluation for Video Generation and Editing, arXiv:2608.05485v1 (6 Aug 2026), CC BY 4.0. Extracted Table 2 T2V and TI2V VideoArgus scores, divided by 10. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2608.05485v1",
    "icon": "arxiv.org",
    "scope": "Authors' own generations and VideoArgus scoring on VideoArgus-Bench. Imported: Table 2 T2V and TI2V leaderboards (two tasks with different inputs). Not imported: TS2V, TV2V, TSV2V leaderboards (subject/video-conditioned generation and editing; few catalogue models). Withheld: Kling v3 Omni (Kling 3.0 Omni / O3 tier not stated), OmniWeaving, Step-Video, SANA-Video (not in catalogue), CogVideoX1.5 in TI2V (I2V checkpoint not in catalogue), LTX-Video 2B (0.9.x version not stated)."
  },
  {
    "action": "new",
    "id": "visual-datapoint-image-bench",
    "name": "Datapoint Image Bench leaderboard (datapointai/text-2-image-human-preferences-2m dataset card)",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/datapointai/text-2-image-human-preferences-2m/blob/e1d8719a2d521eac6c62ee84f329afc2c03ec928/README.md"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "Card front matter at revision e1d8719a2d521eac6c62ee84f329afc2c03ec928: 'license: cc-by-4.0'. Card 'License' section: 'Votes, prompts, and all metadata are released under CC-BY-4.0. The images are outputs of the listed third-party models...' Access-gate prompt: 'You agree to use this dataset under the CC-BY-4.0 license with attribution to Datapoint AI, and you agree not to attempt to re-identify annotators.' Only the public card (and its leaderboard chart) is used; the gated files were not accessed.",
    "attribution": "Datapoint AI. Text-to-Image Human Preferences 2M: the Datapoint Image Bench voting record, Hugging Face dataset datapointai/text-2-image-human-preferences-2m @ e1d8719a2d521eac6c62ee84f329afc2c03ec928, CC BY 4.0. Overall Elo values transcribed from the card's leaderboard chart (board of 2026-08-19); labels mapped only to exact catalogue models. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://huggingface.co/datasets/datapointai/text-2-image-human-preferences-2m/blob/e1d8719a2d521eac6c62ee84f329afc2c03ec928/README.md",
    "icon": "huggingface.co",
    "scope": "Datapoint's own blind human pairwise study (30 T2I models). Only the overall Elo board in the card chart is imported; the trydatapoint.com live leaderboard is not used. The chart's x-axis rank numbers skip 21 and repeat 23, but the 30 printed values are in bar order. Withheld: Grok Imagine (quality; Recraft V4.1 Utility; Vidu Q2; Bria FIBO; Luma UNI 1 Max; ImagineArt 2.0; Cosmos 3 Super; Image-01; Kolors 2.1 (truncated or ambiguous labels - 'Grok Imagine (quality', 'Recraft V4.1 Utility' vs catalogue Utility Pro, 'Luma UNI 1 Max' vs Uni-1.1 Max, 'Cosmos 3 Super' vs Cosmos3-Super-Text2Image, 'Kolors 2.1' vs catalogue Kolors 1.0 - or not in catalogue)."
  },
  {
    "action": "new",
    "id": "visual-datapoint-video-bench",
    "name": "Datapoint Video Bench leaderboard (datapointai/text-to-video-human-preferences-326k dataset card)",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/datapointai/text-to-video-human-preferences-326k/blob/49295b5d625c04480c92c33a2b4c502bf571110b/README.md"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "Card front matter at revision 49295b5d625c04480c92c33a2b4c502bf571110b: 'license: cc-by-4.0'. Card 'License' section: 'Votes, prompts, and all metadata are released under CC-BY-4.0. The videos are outputs of the listed third-party models...' Access-gate prompt: 'You agree to use this dataset under the CC-BY-4.0 license with attribution to Datapoint AI, and you agree not to attempt to re-identify annotators.' Only the public card's leaderboard table is used; gated files not accessed.",
    "attribution": "Datapoint AI. Text-to-Video Human Preferences 326K: the Datapoint Video Bench voting record, Hugging Face dataset datapointai/text-to-video-human-preferences-326k @ 49295b5d625c04480c92c33a2b4c502bf571110b, CC BY 4.0. Extracted the card's Overall Elo table; labels mapped only to exact catalogue models. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://huggingface.co/datasets/datapointai/text-to-video-human-preferences-326k/blob/49295b5d625c04480c92c33a2b4c502bf571110b/README.md",
    "icon": "huggingface.co",
    "scope": "Datapoint's own blind human pairwise study (15 T2V models). Only the card's 'Leaderboard' Elo table imported; the trydatapoint.com live board is not used. Withheld (not in catalogue): MiniMax H3 Max; P-Video-2-Pro; PixVerse V6; Ray 3.2; Pika 2.5."
  },
  {
    "action": "new",
    "id": "visual-evalalign",
    "name": "EvalAlign: Supervised Fine-Tuning Multimodal LLMs with Human-Aligned Data for Evaluating Text-to-Image Models (arXiv:2406.16562v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2406.16562v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2406.16562v3 (latest version v3, 10 Oct 2024) links http://creativecommons.org/licenses/by/4.0/; the v3 HTML header reads 'License: CC BY 4.0 arXiv:2406.16562v3 [cs.CV] 10 Oct 2024'.",
    "attribution": "Zhiyu Tan, Xiaomeng Yang, Luozheng Qin, Mengping Yang, Cheng Zhang, Hao Li (Shanghai Academy of AI for Science / CMU / Fudan). EvalAlign, arXiv:2406.16562v3 (10 Oct 2024), CC BY 4.0. Extracted Table 3 'Human' text-image alignment scores. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2406.16562v3",
    "icon": "arxiv.org",
    "scope": "Authors' own human annotation of images generated by 24 models on 100 prompts. Only Table 3 'Human' (text-image alignment) imported; Table 2 (image faithfulness, same images) and automatic metric columns not imported. Withheld: SDXL Refiner v1.0, LCM LoRA SDXL, Wuerstchen (version unstated), Safe SD variants, Openjourney v2 (ilkerc/openjourney-v2), Redshift, Dreamlike Diffusion v1.0, Vintedois, MultiFusion (not in catalogue)."
  },
  {
    "action": "new",
    "id": "visual-sdxl-lightning",
    "name": "SDXL-Lightning: Progressive Adversarial Diffusion Distillation (arXiv:2402.13929v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2402.13929v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2402.13929v3 (latest version v3, 2 Mar 2024) links http://creativecommons.org/licenses/by/4.0/; the v3 HTML header reads 'License: CC BY 4.0'.",
    "attribution": "Shanchuan Lin, Anran Wang, Xiao Yang (ByteDance). SDXL-Lightning: Progressive Adversarial Diffusion Distillation, arXiv:2402.13929v3 (2 Mar 2024), CC BY 4.0. Extracted Table 2 FID (Whole) for one step setting per catalogue model. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2402.13929v3",
    "icon": "arxiv.org",
    "scope": "Authors' own COCO-10K FID evaluation (ByteDance; SDXL-Lightning rows are lab-reported). Only FID (Whole) imported, one row per catalogue model at its default step count: SDXL 32 steps, LCM 4 steps, SDXL-Turbo 1 step, SDXL-Lightning 4-step UNet. Not imported: LCM 1-step, LCM-LoRA, Turbo 4-step, Lightning 1/2/8-step and LoRA rows (other checkpoints or settings), FID-Patch, CLIP."
  },
  {
    "action": "new",
    "id": "visual-cosmos3-report",
    "name": "Cosmos 3: Omnimodal World Models for Physical AI (arXiv:2606.02800v4)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.02800v4"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2606.02800v4 (latest version v4, 23 Jun 2026) links http://creativecommons.org/licenses/by/4.0/; the v4 HTML header reads 'License: CC BY 4.0'.",
    "attribution": "NVIDIA. Cosmos 3: Omnimodal World Models for Physical AI, arXiv:2606.02800v4 (23 Jun 2026), CC BY 4.0. Extracted Table 11 UniGenBench 'All (1170)' scores, divided by 100. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "evidence_url": "https://arxiv.org/abs/2606.02800v4",
    "icon": "arxiv.org",
    "scope": "NVIDIA's own evaluation (Cosmos rows lab-reported). Only Table 11 UniGenBench 'All' imported; Orig/Phys splits are components, CVTG and image-level metrics (Aesv2, HPSv3) not imported, and the video tables (PAIBench-G, RBench) and Artificial Analysis leaderboard mentions are not used. Withheld: Cosmos3-Super (omnimodel checkpoint, not in catalogue), Cosmos3-Nano (catalogued as a video model; its image row is not mapped). Extended 2026-10-09: Existing entry already matches arxiv.org/html/2606.02800v4; no new match needed."
  },
  {
    "action": "new",
    "id": "visual-mindworldbench",
    "name": "MindWorldBench: Evaluating Mental-State-to-Behavior Reasoning in Image-to-Video Generation (arXiv:2609.39147v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.39147v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2609.39147v1 (v1, 30 Sep 2026) links http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2609.39147v1 [cs.CV] 30 Sep 2026'.",
    "attribution": "Ruiqi Li, Xuanyi Liu et al. (Peking University / USTB). MindWorldBench: Evaluating Mental-State-to-Behavior Reasoning in Image-to-Video Generation, arXiv:2609.39147v1 (30 Sep 2026), CC BY 4.0. Extracted Table 1 Intention Accuracy, divided by 100. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2609.39147v1",
    "icon": "arxiv.org",
    "scope": "Authors' own generations (11 I2V systems) and Gemini 3.1 Pro judging. Only Table 1 Intention Accuracy (the benchmark's target capability) imported; Visual Quality, Commonsense Plausibility and World-State Maintenance columns not imported (same videos). Withheld: Kling (table 'Kling', text 'KlingV3', reference the Kling-Omni report, so Kling 3.0 vs Kling 3.0 Omni is unclear)."
  },
  {
    "action": "new",
    "id": "visual-vigor-bench",
    "name": "ViGoR-Bench: How Far Are Visual Generative Models From Zero-Shot Visual Reasoners? (arXiv:2603.25823v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2603.25823v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2603.25823v1 (v1, 26 Mar 2026, only version) links http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2603.25823v1 [cs.CV] 26 Mar 2026'.",
    "attribution": "Haonan Han et al. (Tsinghua University / Meituan M17 / HKU / SIAT). ViGoR-Bench, arXiv:2603.25823v1 (26 Mar 2026), CC BY 4.0. Extracted Table 3 Result Metric Avg, divided by 100. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2603.25823v1",
    "icon": "arxiv.org",
    "scope": "Authors' own generations and Gemini-2.5-Pro judging. Imported: Table 3 Result Metric Avg, split into an image-model run (without CoT) and a video-generator run. Not imported: Process metrics, CoT rows, SFT/RL ablations (Table 4). Meituan co-authors the report, so LongCat-Image-Edit is lab. Withheld: Step1X-Edit (version unstated), HiDream-E1.1, ICEdit, UniPic2-M-9B, Ovis-U1-3B (not in catalogue), Wan 2.2 (variant unstated), Kling 1.6 (tier unstated)."
  },
  {
    "action": "new",
    "id": "visual-phyground",
    "name": "PhyGround: Benchmarking Physical Reasoning in Generative World Models (arXiv:2605.10806v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.10806v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2605.10806v1 (v1, 11 May 2026) links http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2605.10806v1 [cs.CV] 11 May 2026'.",
    "attribution": "Juyi Lin et al. (Northeastern University / Tulane / University of Washington). PhyGround, arXiv:2605.10806v1 (11 May 2026), CC BY 4.0. Extracted Table 2 human Overall scores (1-5), divided by 5. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2605.10806v1",
    "icon": "arxiv.org",
    "scope": "Authors' own generations and human annotation (352 annotators after QC). Only Table 2 'Overall' imported. Withheld: OmniWeaving, Cosmos-Predict2.5-14B and -2B (not in catalogue)."
  },
  {
    "action": "new",
    "id": "visual-edival",
    "name": "EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (arXiv:2509.13399v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2509.13399v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2509.13399v3 (latest version v3, 17 Mar 2026) links http://creativecommons.org/licenses/by/4.0/; the v3 HTML header reads 'License: CC BY 4.0 arXiv:2509.13399v3 [cs.CV] 17 Mar 2026'.",
    "attribution": "Tianyu Chen, Yasi Zhang et al. (EdiVal-Agent authors). EdiVal-Agent, arXiv:2509.13399v3 (17 Mar 2026), CC BY 4.0. Extracted Table 3 EdiVal-O at turn 3, divided by 100. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "evidence_url": "https://arxiv.org/abs/2509.13399v3",
    "icon": "arxiv.org",
    "scope": "Authors' own multi-turn edits with default settings (Appendix M.5) and automatic scoring. Only Table 3 EdiVal-O at T3 imported (the ranking column). Withheld: Gemini 2.0 Flash (API model id not stated; experimental vs preview image generation unclear), Step1X-Edit (stepfun-ai/Step1X-Edit v1.0 repo, not in catalogue), OmniGen (v1), AnyEdit, UltraEdit, MagicBrush (not in catalogue). Extended 2026-10-09: Scope extension if accepted: Table 3 EdiVal-O at T3 rows 'Step1X-Edit' (-> new step1x-edit-v1-0) and 'OmniGen' (-> new omnigen-v1) now map to catalogue models; remove them from the 'Withheld' list."
  },
  {
    "action": "new",
    "id": "visual-weavebench",
    "name": "WEAVE: Unleashing and Benchmarking the In-context Interleaved Comprehension and Generation (arXiv:2511.11434v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2511.11434v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2511.11434v1 (v1, 14 Nov 2025, only version) links http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0'.",
    "attribution": "Wei Chow, Jiachun Pan et al. (WEAVE authors). WEAVE, arXiv:2511.11434v1 (14 Nov 2025), CC BY 4.0. Extracted Table 2 WEAVEBench Avg for selected editors. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2511.11434v1",
    "icon": "arxiv.org",
    "scope": "Authors' own runs and GPT-4.1 judging. Only Table 2 'Avg' imported for image editors. Withheld: understanding-only VLM rows (Intern3.5-VL, Qwen3-VL, GPT-4o, GPT-4.1), Seedream 4.0 and Nano Banana (two rows each, settings not resolvable from the parsed table), AnyEdit, UltraEdit, VAREdit, OmniGen v1, Ovis-U1, UniPic, UniPic2, NextStep-1, Bagel-Zebra and the authors' WEAVE-100k fine-tune (not in catalogue or not a catalogue checkpoint)."
  },
  {
    "action": "new",
    "id": "visual-knowledge-visualization",
    "name": "Knowledge Visualization: A Benchmark and Method for Knowledge-Intensive Text-to-Image Generation (arXiv:2604.22302v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2604.22302v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2604.22302v1 (v1, 24 Apr 2026, only version) links http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2604.22302v1 [cs.CV] 24 Apr 2026'.",
    "attribution": "Ran Zhao, Sheng Jin, Size Wu, Kang Liao et al. (HUST / HKU / NTU S-Lab). Knowledge Visualization, arXiv:2604.22302v1 (24 Apr 2026), CC BY 4.0. Extracted Table 2 Brief Caption Overall (English), divided by 100. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2604.22302v1",
    "icon": "arxiv.org",
    "scope": "Authors' own generations and MLLM scoring. Only Table 2 'Overall' English column imported. Withheld: 'GPT-Image' (version unstated; cited reference is the DALL-E 3 report), Janus-Pro and show-o2 (sizes unstated), Z-Image (base vs turbo unstated), and the authors' Qwen-Image + KE-Check method."
  },
  {
    "action": "new",
    "id": "visual-hp-edit",
    "name": "HP-Edit: A Human-Preference Post-Training Framework for Image Editing (arXiv:2604.19406v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2604.19406v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2604.19406v1 (v1, 21 Apr 2026, only version) links http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2604.19406v1 [cs.CV] 21 Apr 2026'.",
    "attribution": "Fan Li, Chonghuinan Wang, Lina Lei et al. (Huawei Noah's Ark Lab / HIT / Nankai). HP-Edit, arXiv:2604.19406v1 (21 Apr 2026), CC BY 4.0. Extracted Table 1 RealPref-Bench 'Human' user-study scores, divided by 5. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2604.19406v1",
    "icon": "arxiv.org",
    "scope": "Authors' own edits and five-annotator user study. Only Table 1 'Human' imported; HP-score columns (VLM scorer), GEdit-Bench tables (values match other papers' published numbers) and the authors' HP-Edit model are not imported. Withheld: Step1X-Edit (version unstated), X2Edit (not in catalogue)."
  },
  {
    "action": "new",
    "id": "visual-switch-bench",
    "name": "SWITCH: Benchmarking Modeling and Handling of Tangible Interfaces in Long-horizon Embodied Scenarios (arXiv:2511.17649v5)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2511.17649v5"
    ],
    "exact": true,
    "license": "CC-BY-SA-4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2511.17649v5 (latest version v5, 7 Jul 2026) links http://creativecommons.org/licenses/by-sa/4.0/; the v5 HTML header reads 'License: CC BY-SA 4.0'. ShareAlike applies to the adapted numerical material.",
    "attribution": "Juntao Cheng, Wanyue Zhang et al. (BAAI, SWITCH authors). SWITCH, arXiv:2511.17649v5 (7 Jul 2026), CC BY-SA 4.0. Extracted Table 3 single-image-to-video VLM-as-Judge Overall scores, divided by 5. This adaptation is licensed CC BY-SA 4.0. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2511.17649v5",
    "icon": "arxiv.org",
    "scope": "Authors' own generations and Gemini 3.1 Pro judging (generative track only; the MCQ/OpenQA tables evaluate understanding models and are not used). Only the I2V VLM Overall column imported. Withheld: Wan 2.2 (variant not stated). The HF Space leaderboard is not used."
  },
  {
    "action": "new",
    "id": "visual-wrop-object-permanence",
    "name": "Training Object Permanence in World Models (WROP exam; arXiv:2609.28654v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.28654v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://creativecommons.org/licenses/by/4.0/",
    "quote": "arXiv abs page https://arxiv.org/abs/2609.28654v1 (v1, 23 Sep 2026, only version) links http://creativecommons.org/licenses/by/4.0/; the HTML header reads 'License: CC BY 4.0 arXiv:2609.28654v1 [cs.AI] 23 Sep 2026'.",
    "attribution": "Haotian Zhang, Fengyuan Yu et al. (USC / CMU / UMich / JHU / UCSD). Training Object Permanence in World Models, arXiv:2609.28654v1 (23 Sep 2026), CC BY 4.0. Extracted Table 2 human Elo. No values estimated or rerun. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "evidence_url": "https://arxiv.org/abs/2609.28654v1",
    "icon": "arxiv.org",
    "scope": "Authors' own generations and blind human pairwise study. Only Table 2 overall Elo imported; per-family Elo (Table 6) and automatic metrics (Table 3) not imported. Withheld: PWM-WROP (authors' model), Wan 3.0 Prime (tier not in catalogue), Runway Aleph 2 (not catalogued; catalogue has Aleph), Grok Imagine (video extend) (version unstated), LTX-2.3 Extend and LTX-2.3 Dev (two interface variants of one model), Cosmos3 Super, MAGI-1, Wan-VACE, HY-OmniWeaving (not in catalogue)."
  },
  {
    "id": "first-party-rev",
    "name": "Rev AI official pricing and documentation",
    "match": [
      "rev.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://www.rev.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-10-07",
    "exact": true,
    "icon": "rev.ai",
    "attribution": "Rev AI official pricing page, observed 2026-10-07",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "first-party-prime-rl",
    "name": "PRIME-RL official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/PRIME-RL/Eurus-2-7B-PRIME/blob/2f47d199012a5d37d0dcedbeff1e2802a0ea443c/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/PRIME-RL/Eurus-2-7B-PRIME/blob/2f47d199012a5d37d0dcedbeff1e2802a0ea443c/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "PRIME-RL official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-07"
  },
  {
    "id": "first-party-berkeley-nest",
    "name": "Berkeley NEST official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/berkeley-nest/Starling-LM-7B-alpha/blob/1dddf3b95bc1391f6307299eb1c162c194bde9bd/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha/blob/1dddf3b95bc1391f6307299eb1c162c194bde9bd/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Berkeley NEST official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-07"
  },
  {
    "id": "first-party-nexusflow",
    "name": "Nexusflow official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/Nexusflow/Athene-V2-Chat/blob/493f1bbd561a5a7e3d27c4081d4ee47508bf6831/config.json",
      "huggingface.co/Nexusflow/Starling-LM-7B-beta/blob/39a4d501472dfede947ca5f4c5af0c1896d0361b/config.json",
      "huggingface.co/Nexusflow/NexusRaven-V2-13B/blob/a6b84dc52f4c8f2cd61637f840066f376c629271/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/Nexusflow/Athene-V2-Chat/blob/493f1bbd561a5a7e3d27c4081d4ee47508bf6831/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Nexusflow official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-upstage",
    "name": "Upstage official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0/blob/090d405e6fe7b4c7b0b999a67f5c6dee0514a7ad/config.json",
      "huggingface.co/upstage/Solar-Open2-250B/blob/9190fbe63a2ad8e17fc766ccceb36de7c66f004b/README.md",
      "huggingface.co/upstage/solar-pro-preview-instruct/blob/dd4bcf7006df9b1ce3f87711e702e4063832aae3/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0/blob/090d405e6fe7b4c7b0b999a67f5c6dee0514a7ad/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Upstage official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-internlm",
    "name": "InternLM (Shanghai AI Lab) official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/internlm/internlm2_5-20b-chat/blob/13741228cfb01d3a69c8f83169a105de77286b8a/config.json",
      "huggingface.co/internlm/internlm2_5-1_8b-chat/blob/57dcbc5455bf050d27a80f0ebd7808387caa714d/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/internlm/internlm2_5-20b-chat/blob/13741228cfb01d3a69c8f83169a105de77286b8a/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "InternLM (Shanghai AI Lab) official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-llava",
    "name": "LLaVA official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/liuhaotian/llava-v1.6-34b/blob/6754aa86582435ae12ef47db2de51180f0968aa7/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/liuhaotian/llava-v1.6-34b/blob/6754aa86582435ae12ef47db2de51180f0968aa7/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "LLaVA official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-07"
  },
  {
    "id": "first-party-phind",
    "name": "Phind official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/Phind/Phind-CodeLlama-34B-v2/blob/949f61e203f91b412efe8f679c798f09f0ff4b0c/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/Phind/Phind-CodeLlama-34B-v2/blob/949f61e203f91b412efe8f679c798f09f0ff4b0c/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Phind official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-07"
  },
  {
    "id": "first-party-wizardlm",
    "name": "WizardLM official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/WizardLM/WizardCoder-33B-V1.1/blob/22d03e16bd052cec88228e17e832f851391e3565/config.json",
      "huggingface.co/WizardLM/WizardCoder-Python-13B-V1.0/blob/5ac6748b1f5a4c282107ddc7d3b69fdc4a686d75/config.json",
      "huggingface.co/WizardLM/WizardCoder-Python-34B-V1.0/blob/897fc6d9e12136c68c441b2350d015902c144b20/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/WizardLM/WizardCoder-33B-V1.1/blob/22d03e16bd052cec88228e17e832f851391e3565/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "WizardLM official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-07"
  },
  {
    "id": "first-party-open-thoughts",
    "name": "Open Thoughts official model cards and configuration",
    "kind": "first-party",
    "match": [
      "huggingface.co/open-thoughts/OpenThinker3-7B/blob/864fb9aefbb91f257c4911e4b9f85a6986f8eab2/config.json",
      "huggingface.co/open-thoughts/OpenThinker2-32B/blob/0aefafe3ee27e073e14fb94a0563795fddf38c50/config.json",
      "huggingface.co/open-thoughts/OpenThinker-7B/blob/4ba9dee9b7e6e4796c6bccf6e15c2f5b60843e1e/config.json",
      "huggingface.co/open-thoughts/OpenThinker-32B/blob/840749ca114a8f02be56f378d2999ad6d9e42d4f/config.json",
      "huggingface.co/open-thoughts/OpenThinker2-7B/blob/57922a00cca8bae9776277bc51c3aceedacc643c/config.json"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/open-thoughts/OpenThinker3-7B/blob/864fb9aefbb91f257c4911e4b9f85a6986f8eab2/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Open Thoughts official model cards and configuration, observed 2026-10-07",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "acl-2021-findings-acl-404-prost",
    "name": "PROST (Aroca-Ouellette et al., Findings of ACL-IJCNLP 2021) - authors' own zero-shot runs",
    "kind": "paper",
    "match": [
      "aclanthology.org/2021.findings-acl.404.pdf"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "terms_url": "https://aclanthology.org/2021.findings-acl.404/",
    "license_url": "/licenses/_cc-by-4.0-legalcode.txt",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "quote": "ACL Anthology landing page footer for 2021.findings-acl.404: 'Materials published in or after 2016 are licensed on a Creative Commons Attribution 4.0 International License.' Findings of ACL-IJCNLP 2021 is an ACL publication. Section 4: 'We use Huggingface's (Wolf et al., 2020) pretrained models, see Table 4 for the full set.'",
    "attribution": "Stephane Aroca-Ouellette, Cory Paik, Alessandro Roncone and Katharina Kann; (c)2021 Association for Computational Linguistics. PROST: Physical Reasoning about Objects through Space and Time. Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4597-4608.\nSource: https://aclanthology.org/2021.findings-acl.404.pdf\nLicence: CC BY 4.0 - https://creativecommons.org/licenses/by/4.0/\nChanges: selected numerical cells (Table 5 macro averages) transcribed into JSON; percentages divided by 100; model identities clarified. No questions or figures redistributed. No endorsement.",
    "scope": "Table 5 'Macro Average' column for GPT (openai-gpt) and GPT-2 B/M/L/XL, the five catalogue models. BERT/RoBERTa/ALBERT rows (not in catalogue), T5 and UnifiedQA rows (9-concept average without sliding; UnifiedQA fine-tuned), per-concept columns, Tables 6-7 not imported.",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-cohere-scores",
    "name": "Cohere vendor-reported benchmark scores for its own models (cohere.com blog)",
    "kind": "first-party",
    "match": [
      "cohere.com/blog/command-a-plus"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://cohere.com/terms-of-use",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "cohere.com",
    "attribution": "Cohere, vendor-reported results for Cohere's own models from the 'Introducing Command A+' blog post (May 20, 2026); individual figures from the post's text (2026-10-08) and printed chart labels (2026-10-09) hand-recorded; percentages converted to fractions. Lab-reported, not independent measurements. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07 (UPDATING.md 'Vendor-reported scores'): a vendor's own published score for its own model, from its announcement, system/model card or docs, may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells and cells credited to a third party excluded. The post's Artificial Analysis Intelligence Index (37) is credited to Artificial Analysis and is not recorded; MMMU / MMMU Pro / MathVista figures are not recorded because the post gives no split (val/test, testmini, standard/vision).",
    "scope": "Command A+ (05-2026) with its printed comparators Command A Reasoning (08-2025) and Command A Vision (07-2025). Body text: τ²-Bench Telecom 85% vs 37%, Terminal-Bench Hard 25% vs 3%, CharXiv reasoning 52.7% vs 46.9%. Chart data labels (Image 3 and Image 5, values printed on the charts, settings in footnote 1): AIME 2025 90% vs 57%, IFBench (loose, prompt-level) 74% vs 36%, SciCode (level unspecified) 38% vs 30%, CharXiv descriptive 88% vs 82%. Not recorded: MT-AIME (internal translation) and WMT24++ (includes internal translations), internal North evaluations, and the Artificial Analysis Intelligence Index."
  },
  {
    "id": "lmsys-arena-space-mtbench",
    "name": "LMSYS Chatbot Arena leaderboard Space: MT-Bench column of the published leaderboard table",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/lmarena-ai/arena-leaderboard/blob/6f2eeeba09bf42144dbc118fe1a8e12e4c533f9d/leaderboard_table_20250804.csv"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/spaces/lmarena-ai/arena-leaderboard/blob/6f2eeeba09bf42144dbc118fe1a8e12e4c533f9d/README.md",
    "quote": "Space README front matter at the pinned revision 6f2eeeb: 'license: apache-2.0' (public, not gated). The CSV is a committed file of the Space repository. Apache-2.0 §2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce and distribute the Work and Derivative Works.",
    "attribution": "LMSYS / LMArena, Chatbot Arena leaderboard Space (lmarena-ai/arena-leaderboard, formerly lmsys/chatbot-arena-leaderboard), leaderboard_table_20250804.csv @6f2eeeb, Apache-2.0. MT-Bench column only; Arena Elo and MMLU columns not used.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://huggingface.co/spaces/lmarena-ai/arena-leaderboard/blob/6f2eeeba09bf42144dbc118fe1a8e12e4c533f9d/README.md",
    "icon": "huggingface.co",
    "scope": "MT-Bench column, rows first published in the 2023 tables (LMSYS's own fastchat.llm_judge runs). This is the Hugging Face Space git repository, not the arena.ai website; models.json identity notes already cite this Space. Review flag: the owner should confirm the Space's Apache-2.0 tag is acceptable for the table data (it is LMSYS's own measurement, so no third-party permission is needed for these rows)."
  },
  {
    "id": "paper-2408.12570v1-jamba15",
    "name": "Jamba-1.5: Hybrid Transformer-Mamba Models at Scale (AI21, arXiv 2408.12570v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2408.12570v1"
    ],
    "license": "CC BY-SA 4.0",
    "license_url": "http://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://arxiv.org/abs/2408.12570v1",
    "quote": "The arXiv abstract page for 2408.12570 links the licence 'http://creativecommons.org/licenses/by-sa/4.0/' in its licence field (only version v1 exists).",
    "attribution": "Jamba Team (AI21 Labs), 'Jamba-1.5: Hybrid Transformer-Mamba Models at Scale', arXiv:2408.12570v1, CC BY-SA 4.0. Values transcribed from Tables 2-3 and divided by 100; share-alike applies to adapted tables.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2408.12570v1",
    "icon": "arxiv.org",
    "scope": "Jamba-1.5 Mini/Large rows (AI21's own runs, by=lab) and competitor cells marked '† evaluation run by us' (by=independent). Unmarked competitor cells (reported elsewhere) and ⋄ cells (Open LLM Leaderboard) excluded. Preferred over the gated HF card (jamba-open-model-license), whose table is a later CoT re-run with different values."
  },
  {
    "id": "paper-2410.07073v2-pixtral",
    "name": "Pixtral 12B (Mistral AI, arXiv 2410.07073v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2410.07073v2"
    ],
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2410.07073v2",
    "quote": "The arXiv abstract page for 2410.07073 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field. Table 2 caption: 'We re-evaluate all models with the same prompt and evaluation metric'; Sec. 4.1: 'Table 3 evaluates Pixtral 12B against open-source models of comparable size on common text-only benchmarks (again, with common prompting and evaluation protocols)'.",
    "attribution": "Agrawal et al. (Mistral AI), 'Pixtral 12B', arXiv:2410.07073v2, CC BY 4.0. Values transcribed from Tables 2-3; percentages divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2410.07073v2",
    "icon": "arxiv.org",
    "scope": "Mistral's own re-evaluation of every listed model (Pixtral 12B rows by=lab, others by=independent). LMSys-Vision column (copied Arena Elo), grey flexible-parsing rows, DocVQA (validation split; the existing docvqa id is the test split) and Table 3 MT-Bench (judge not stated) excluded. Not mapped: LLaVA-OneVision 7B/72B (ov / ov-sft / ov-chat checkpoint not stated), Gemini-1.5-Flash 8B (0827) (experimental checkpoint, not -001)."
  },
  {
    "id": "molmo-card-human-elo",
    "name": "Ai2 Molmo-72B-0924 model card: Human Preference Elo Rating column",
    "kind": "first-party",
    "match": [
      "huggingface.co/allenai/Molmo-72B-0924/blob/ab586dd9640de059566975b8e5bcd0c6fdc73956/README.md#evaluations"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/allenai/Molmo-72B-0924/blob/ab586dd9640de059566975b8e5bcd0c6fdc73956/README.md",
    "terms_url": "https://huggingface.co/allenai/Molmo-72B-0924/blob/ab586dd9640de059566975b8e5bcd0c6fdc73956/README.md",
    "quote": "Model card front matter at the pinned revision: 'license: apache-2.0' (repository licence, not gated). The Elo column is Ai2's own human study (Molmo paper, arXiv 2409.17146, Sec. 'Human evaluation', describes Ai2 querying all 27 models and collecting the pairwise judgments), so Ai2 is the measurement source for every row.",
    "attribution": "Allen Institute for AI (Ai2), Molmo-72B-0924 model card @ab586dd, Apache-2.0. 'Human Preference Elo Rating' column only.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2409.17146",
    "icon": "huggingface.co",
    "scope": "Human Preference Elo column only. 'Average Score on 11 Academic Benchmarks' not used (the paper mixes reported and re-run numbers for other models). Fragment-scoped so any other use of this card stays separate. Not mapped: GPT-4o and GPT-4V (snapshot not stated), Gemini 1.5 Pro/Flash (001 vs 002 not stated), InternVL2 8B/Llama-76B, LLaVA-OneVision (ov/ov-sft/ov-chat not stated), Cambrian-1, xGen-MM, PaliGemma, LLaVA-1.5, Molmo 7B-O and MolmoE 1B (not in the catalogue)."
  },
  {
    "id": "paper-2601.10611v4-molmo2",
    "name": "Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding (Ai2, arXiv 2601.10611v4)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2601.10611v4"
    ],
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2601.10611v4",
    "quote": "The arXiv abstract page for 2601.10611 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field.",
    "attribution": "Ai2 (Allen Institute for AI), 'Molmo2: Open Weights and Data for Vision-Language Models with Video Understanding and Grounding', arXiv:2601.10611v4, CC BY 4.0. Values transcribed from Tables 2 and 6; percentages divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2601.10611v4",
    "icon": "arxiv.org",
    "scope": "Only Ai2-measured values: the human-preference Elo, Molmo2-CapTest F1 and Molmo2-VideoCount columns of Table 2 (Ai2's own study / new Ai2 benchmarks, all rows), and Ai2's own Molmo/Molmo2 rows in Table 6. Other competitor benchmark cells are excluded because Sec. 4.1 says 'We prioritize author-published results but fill in missing results with the best previously reported values'. Not mapped (not in the catalogue): Qwen3-VL-4B/8B, InternVL3.5, Keye-VL, GLM-4.1V, MiniCPM-V-4.5, Eagle2.5, PLM, LLaVA-Video, VideoChat-Flash, Molmo2-4B/O-7B."
  },
  {
    "id": "paper-2404.14219v3-phi3",
    "name": "Phi-3 Technical Report (Microsoft, arXiv 2404.14219v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2404.14219v3"
    ],
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2404.14219v3",
    "quote": "The arXiv HTML of v3 prints 'License: CC BY 4.0' and the abstract page links http://creativecommons.org/licenses/by/4.0/. Table 2 caption: 'All the reported numbers are produced with the exact same pipeline to ensure that the numbers are comparable except for MM1-3B-Chat and MM1-7B-Chat'.",
    "attribution": "Microsoft, 'Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone', arXiv:2404.14219v3, CC BY 4.0. Values from Table 2 divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2404.14219v3",
    "icon": "arxiv.org",
    "scope": "Phi-3-Vision table (v3, May 2024) only: Microsoft-run rows for Phi-3-Vision (lab), Claude 3 Haiku and GPT-4V-Turbo (2024-04-09). MM1 columns (copied), LLaVA-1.6 / LLaVA-Next / Qwen-VL-Chat (not in the catalogue), Gemini 1.0 Pro V (vision model not in the catalogue), MMBench (version not stated), ScienceQA/Inter-GPS/POPE (no ids) excluded."
  },
  {
    "id": "microsoft-phi35-vision-card-benchmarks",
    "name": "Microsoft Phi-3.5-vision-instruct model card: Benchmarks table",
    "kind": "first-party",
    "match": [
      "huggingface.co/microsoft/Phi-3.5-vision-instruct/blob/12b77fb40b63a2c73c68243d3f767aab688a1b2a/README.md#benchmarks"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/microsoft/Phi-3.5-vision-instruct/blob/12b77fb40b63a2c73c68243d3f767aab688a1b2a/LICENSE",
    "terms_url": "https://huggingface.co/microsoft/Phi-3.5-vision-instruct/blob/12b77fb40b63a2c73c68243d3f767aab688a1b2a/README.md",
    "quote": "Card front matter at the pinned revision: 'license: mit'; card section 'License': 'The model is licensed under the MIT license.' The table is Microsoft's own run: 'we compare Phi-3.5-vision with a set of models over a variety of zero-shot benchmarks using our internal benchmark platform'.",
    "attribution": "Microsoft, Phi-3.5-vision-instruct model card @12b77fb (MIT). Values divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://huggingface.co/microsoft/Phi-3.5-vision-instruct/blob/12b77fb40b63a2c73c68243d3f767aab688a1b2a/README.md#benchmarks",
    "icon": "huggingface.co",
    "scope": "Benchmarks table, Microsoft-run. Not mapped: Intern-VL-2-8B (not in the catalogue), Gemini-1.5-Flash and Gemini-1.5-Pro (snapshot not stated; Aug 2024 had stable and experimental versions). MMBench (version not stated), ScienceQA, InterGPS and POPE (no ids) excluded."
  },
  {
    "id": "first-party-deepseek-scores",
    "name": "DeepSeek vendor-reported benchmark scores for its own models (Hugging Face model cards)",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V2.5-1210/blob/6f134cbe88cb9284a8ce696e8ac8eefd0bc24ede/README.md#1-introduction",
      "huggingface.co/deepseek-ai/DeepSeek-V2-Lite-Chat/blob/85864749cd611b4353ce1decdb286193298f64c7/README.md#chat-model"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V2.5-1210/blob/6f134cbe88cb9284a8ce696e8ac8eefd0bc24ede/README.md",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'.",
    "attribution": "DeepSeek, vendor-reported results for DeepSeek's own models from the pinned DeepSeek-V2.5-1210 card; hand-recorded 2026-10-08; percentages converted to fractions. Lab-reported. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "huggingface.co",
    "scope": "MATH-500 sentence of the Introduction only (both figures are DeepSeek's own models). The LiveCodeBench (08.01-12.01) pair is not recorded: no existing benchmark id covers that window. Fragment-scoped so it stays separate from any price/context entry for the same card."
  },
  {
    "id": "paper-2406.12793v2-chatglm",
    "name": "ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools (Team GLM, arXiv 2406.12793v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2406.12793v2"
    ],
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2406.12793v2",
    "quote": "The arXiv HTML of v2 prints 'License: CC BY 4.0' and the abstract page links http://creativecommons.org/licenses/by/4.0/. Sec. 3.6: 'To evaluate the performance of GLM models on function call, we carry out evaluations on Berkeley Function Call Leaderboard'.",
    "attribution": "Team GLM (Zhipu AI / Tsinghua), 'ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools', arXiv:2406.12793v2, CC BY 4.0. Table 7 Overall values divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2406.12793v2",
    "icon": "arxiv.org",
    "scope": "Table 7 (BFCL Overall, authors' runs) only. GLM-4-Air (0605) not in the catalogue. Table 1's ChatGLM3 column is ChatGLM3-6B-Base (not the catalogue chat model), so Table 1 was not used for chatglm3-6b. Pass 12 (2026-10-11): also Table 2 'GLM-4 performance on academic benchmarks', GLM-4 (0116) and GLM-4 (0520) rows only (Zhipu's own models, by: lab; MMLU, GSM8K, MATH, BBH, GPQA, HumanEval); baseline rows are not the authors' runs and are not used."
  },
  {
    "id": "first-party-meta-asr-cards",
    "name": "Meta (facebook org) Hugging Face ASR model cards: published LibriSpeech WER for its own checkpoints",
    "kind": "first-party",
    "match": [
      "huggingface.co/facebook/data2vec-audio-base-960h/blob/32331f3123e703528918aa688a9a38232d58c872/README.md#evaluation",
      "huggingface.co/facebook/data2vec-audio-large-960h/blob/27aba26eed532b86dcd0f17284a0307de4b51f39/README.md#evaluation",
      "huggingface.co/facebook/wav2vec2-conformer-rope-large-960h-ft/blob/6b36ef01c6443c67ae7ed0822876d091ab50e4aa/README.md#evaluation",
      "huggingface.co/facebook/wav2vec2-conformer-rel-pos-large-960h-ft/blob/ca7f36f527f234b3cd4f05ecee30361f971e8e33/README.md#evaluation",
      "huggingface.co/facebook/wav2vec2-base-960h/blob/22aad52d435eb6dbaf354bdad9b0da84ce7d6156/README.md#evaluation",
      "huggingface.co/facebook/wav2vec2-large-960h/blob/bdeaacdf88f7a155f50a2704bc967aa81fbbb2ab/README.md#evaluation",
      "huggingface.co/facebook/wav2vec2-large-960h-lv60-self/blob/54074b1c16f4de6a5ad59affb4caa8f2ea03a119/README.md#evaluation",
      "huggingface.co/facebook/hubert-large-ls960-ft/blob/ece5fabbf034c1073acae96d5401b25be96709d8/README.md#evaluation",
      "huggingface.co/facebook/hubert-xlarge-ls960-ft/blob/86a09e67e0c8d074533992379242405825516eca/README.md#evaluation"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/facebook/wav2vec2-base-960h/blob/22aad52d435eb6dbaf354bdad9b0da84ce7d6156/README.md",
    "terms_url": "https://huggingface.co/facebook",
    "quote": "Each card's front matter at the pinned revision declares 'license: apache-2.0' (repository licence, not gated). The WER figures are the maker's own published results for its own checkpoints (evaluation section 'Result (WER)' and model-index), so the vendor-reported rule also applies.",
    "attribution": "Meta AI (facebook organisation on Hugging Face), model cards for data2vec-audio, wav2vec2(-conformer) and HuBERT LibriSpeech checkpoints at the pinned revisions, Apache-2.0. WER % divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "huggingface.co",
    "scope": "LibriSpeech test-clean/test-other WER printed on nine pinned facebook/* ASR cards, pooled as one maker's evaluation (same greedy transformers snippet), like the existing pooled vendor entries. Each card alone scores one checkpoint, which is why the first pass could not form a run from them. wav2vec2-large-robust-ft-libri-960h prints no WER; not used."
  },
  {
    "id": "paper-2410.05677v3-t2vturbo2",
    "name": "T2V-Turbo-v2: Enhancing Video Generation Model Post-Training through Data, Reward, and Conditional Guidance Design (arXiv 2410.05677v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2410.05677v3"
    ],
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2410.05677v3",
    "quote": "The arXiv abstract page for 2410.05677 links the licence 'http://creativecommons.org/licenses/by/4.0/' in its licence field.",
    "attribution": "Li et al., 'T2V-Turbo-v2', arXiv:2410.05677v3, CC BY 4.0. VBench Total Score values divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2410.05677v3",
    "icon": "arxiv.org",
    "scope": "Table 1 Total Score for the authors' own T2V-Turbo (VC2) and T2V-Turbo-v2 (w/ MG) only; Quality/Semantic sub-scores (components of the total) and competitor columns (VBench-leaderboard copies) excluded."
  },
  {
    "id": "first-party-wizardlm-scores",
    "name": "WizardLM team vendor-reported scores for its own WizardCoder models (Hugging Face card)",
    "kind": "first-party",
    "match": [
      "huggingface.co/WizardLMTeam/WizardCoder-Python-13B-V1.0/blob/5ac6748b1f5a4c282107ddc7d3b69fdc4a686d75/README.md#wizardcoder-empowering-code-large-language-models-with-evol-instruct"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/WizardLMTeam/WizardCoder-Python-13B-V1.0/blob/5ac6748b1f5a4c282107ddc7d3b69fdc4a686d75/README.md",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'.",
    "attribution": "WizardLM team, WizardCoder family table in the pinned WizardCoder-Python-13B-V1.0 card; HumanEval pass@1 hand-recorded 2026-10-08; percentages converted to fractions. Lab-reported. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "huggingface.co",
    "scope": "HumanEval column for the three catalogue WizardCoder models (Python-13B, Python-34B, 33B-V1.1). The card licence is llama2 (not an open data licence), so only the vendor rule applies."
  },
  {
    "id": "primeintellect-intellect3-card-eval",
    "name": "Prime Intellect INTELLECT-3 model card: Evaluation table (Environments Hub runs)",
    "kind": "first-party",
    "match": [
      "huggingface.co/PrimeIntellect/INTELLECT-3/blob/ff39d4a4688989f3f28868923d030c28e1b7d81c/README.md#evaluation"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/PrimeIntellect/INTELLECT-3/blob/ff39d4a4688989f3f28868923d030c28e1b7d81c/README.md",
    "terms_url": "https://huggingface.co/PrimeIntellect/INTELLECT-3/blob/ff39d4a4688989f3f28868923d030c28e1b7d81c/README.md",
    "quote": "Card front matter at the pinned revision: 'license: mit'; card text: 'The model, training frameworks, and environments are open-sourced under fully-permissive licenses (MIT and Apache 2.0).' Provenance (technical report, https://storage.googleapis.com/intellect-3-paper/INTELLECT_3_Technical_Report.pdf, Fig. 1 note): 'All the models above were evaluated using our public, reproducible Environments Hub implementations.'",
    "attribution": "Prime Intellect, INTELLECT-3 model card @ff39d4a (MIT). Values divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://storage.googleapis.com/intellect-3-paper/INTELLECT_3_Technical_Report.pdf",
    "icon": "huggingface.co",
    "scope": "Evaluation table (Prime Intellect's own runs of all listed models). Excluded: GLM-4.5 MMLU-Pro (the report marks it 'Reported by AA Index', i.e. Artificial Analysis, which the project does not use). The report PDF itself states no licence, so the MIT card is the cited copy."
  },
  {
    "id": "speech-paradee-report-ccby",
    "name": "Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice TTS Model (arXiv:2610.06817v1), Table 3 small-model comparison",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2610.06817v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2610.06817v1",
    "quote": "arXiv abstract page for 2610.06817v1 links http://creativecommons.org/licenses/by/4.0/ and the HTML header states 'License: CC BY 4.0'. Table 3 caption: 'Paradee against other small open models on the same 200 held-out sentences, with the same scoring. ... Speed is real-time factor on one CPU thread of the same machine.' §5: 'All UTMOS scores are averages over the 200 held-out sentences.' 'Word error rate (WER) comes from transcribing the audio with Whisper (base)'. The single author ran every row.",
    "attribution": "Sahil Mahendrakar, 'Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text-to-Speech Model', arXiv:2610.06817v1 (2026), CC BY 4.0. Table 3 UTMOS and WER, UTMOS stored (x-1)/4 and WER % / 100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2610.06817v1",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-speechbrain-asr-cards",
    "name": "SpeechBrain Hugging Face ASR model cards: published LibriSpeech/VoxPopuli WER for its own checkpoints",
    "kind": "first-party",
    "match": [
      "huggingface.co/speechbrain/asr-wav2vec2-librispeech/blob/356e347b73885161e8996d9a79b6588dc8bf9f9b/README.md",
      "huggingface.co/speechbrain/asr-conformer-loquacious/blob/42106e67e454ae361ca2d8311a7a807d5f95a3b9/README.md",
      "huggingface.co/speechbrain/asr-transformer-transformerlm-librispeech/blob/bcf56f762f37877912578f8800bc2a4f5171e5e5/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/speechbrain/asr-wav2vec2-librispeech/blob/356e347b73885161e8996d9a79b6588dc8bf9f9b/README.md",
    "terms_url": "https://huggingface.co/speechbrain",
    "quote": "Each card's front matter at the pinned revision declares 'license: apache-2.0' (repository licence, not gated). The WER figures are SpeechBrain's own published results for its own checkpoints ('The performance of the model is the following:' tables and model-index), so the vendor-reported rule also applies.",
    "attribution": "SpeechBrain (speechbrain organisation on Hugging Face), model cards for asr-wav2vec2-librispeech, asr-conformer-loquacious (Samsung AI Center-Cambridge contribution) and asr-transformer-transformerlm-librispeech at the pinned revisions, Apache-2.0. WER % divided by 100.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "huggingface.co",
    "scope": "LibriSpeech test-clean/test-other and VoxPopuli test WER printed on three pinned speechbrain/* cards, pooled as one maker's evaluation like first-party-meta-asr-cards. PERFORMANCE.md (recipe-level numbers that can differ from the released checkpoint, e.g. wav2vec2 1.65/3.67 vs card 1.90/3.96) is not used."
  },
  {
    "id": "speech-carun-tts-bench-apache",
    "name": "carun/tts-bench: intelligibility benchmark of four open TTS engines (published run, 2026-09-10)",
    "kind": "dataset",
    "match": [
      "github.com/carun/tts-bench/blob/27727b24abbc688f718992eb5bfc5c7121facfcb/data/REPORT_DATA.md",
      "github.com/carun/tts-bench/blob/27727b24abbc688f718992eb5bfc5c7121facfcb/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/carun/tts-bench/blob/27727b24abbc688f718992eb5bfc5c7121facfcb/LICENSE.txt",
    "terms_url": "https://github.com/carun/tts-bench/blob/27727b24abbc688f718992eb5bfc5c7121facfcb/LICENSE.txt",
    "quote": "Repository LICENSE.txt at the pinned commit is the unmodified Apache License 2.0 (GitHub reports spdx Apache-2.0); no carve-out for data. Commit 27727b2 'data: raw results from the published run' adds data/REPORT_DATA.md and per-utterance JSON. README: 'Two independent ASR systems then transcribe every output, and word error rate is computed against the input text after identical normalisation of both sides.' The author ran all four engines.",
    "attribution": "Arun C (carun), tts-bench, github.com/carun/tts-bench @27727b2, Apache-2.0. Whisper-judge overall WER, % / 100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "github.com"
  },
  {
    "id": "paper-2610.09823v1-ultratext",
    "name": "UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation (arXiv:2610.09823v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2610.09823v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2610.09823v1",
    "quote": "arXiv abstract page for 2610.09823v1 links 'http://creativecommons.org/licenses/by/4.0/' as the licence; the HTML version prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Deyuan Liu, Yihao Hu, Jingxuan Zhang, Xingying Li, Jun Xie, Jiacheng Liu, Jungang Li, Yu Huang, Xuanyi Liu, Yue Ding, Zecheng Wang, Lei Zhao, Mingda Wang, Zhenglin Cheng, Peng Sun, Tao Lin. UltraText Bench, arXiv:2610.09823v1, CC BY 4.0. Table 5 Composite column extracted; percentages divided by 100; model labels mapped to catalogue ids as noted per row. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2610.09823v1",
    "icon": "arxiv.org",
    "scope": "Table 5 Composite column only, for the 24 configurations the authors evaluated themselves (Sec. 5.1 'We evaluated 24 model configurations'). Boogu-Image and HiDream-O1-Image-Dev rows not catalogued; sub-dimension and per-level columns not used (correlated slices)."
  },
  {
    "id": "paper-2603.26174v1-creval",
    "name": "CREval: An Automated Interpretable Evaluation for Creative Image Manipulation under Complex Instructions (arXiv:2603.26174v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2603.26174v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2603.26174v1",
    "quote": "arXiv abstract page for 2603.26174v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Chonghuinan Wang, Hongxun Yao et al. (Harbin Institute of Technology), CREval, arXiv:2603.26174v1, CC BY 4.0. Table 2 Overall avg column extracted; divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2603.26174v1",
    "icon": "arxiv.org",
    "scope": "Table 2 Overall avg for the models the authors evaluated ('we systematically evaluate a diverse set of state-of-the-art open and closed-source models'). ICEdit not catalogued; 'Qwen-image-Edit' row skipped because the text calls the same reference Qwen-Image-Edit-2509; '(think)' rows skipped."
  },
  {
    "id": "paper-2604.24023v2-servimage",
    "name": "ServImage: An Image Generation and Editing Benchmark from Real-world Commercial Imaging Services (arXiv:2604.24023v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2604.24023v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2604.24023v2",
    "quote": "arXiv abstract page for 2604.24023v2 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Fengxian Ji et al., ServImage, arXiv:2604.24023v2, CC BY 4.0. Table 2 Task Acc. column extracted; divided by 100; model labels mapped via the Appendix Table 6 API endpoints. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2604.24023v2",
    "icon": "arxiv.org",
    "scope": "Table 2 'Task Acc. (%)' (human payment-decision labels) for 12 of the 16 models the authors ran. Not used: Gemini-Banana-Pro (endpoint 'gemini-2.5-flash-image-pro' at $0.028/call does not identify Nano Banana Pro), Kling-v2 images and Midjourney relax modes (not catalogued / version unstated); revenue, share and ServImageModel-predicted columns."
  },
  {
    "id": "card-opengvlab-internvl2-cards",
    "name": "OpenGVLab InternVL2 model cards (InternVL2-4B and InternVL2-26B), Evaluation tables",
    "kind": "first-party",
    "match": [
      "huggingface.co/OpenGVLab/InternVL2-26B/blob/46f37972e05604cc44a59f72334df220fa197a7b/README.md#evaluation",
      "huggingface.co/OpenGVLab/InternVL2-4B/blob/6eb21ec278fbdb03b751a3d48d8dfc9724e05608/README.md#evaluation"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/OpenGVLab/InternVL2-26B/blob/46f37972e05604cc44a59f72334df220fa197a7b/README.md",
    "terms_url": "https://huggingface.co/OpenGVLab/InternVL2-26B/blob/46f37972e05604cc44a59f72334df220fa197a7b/README.md",
    "quote": "Both pinned READMEs declare 'license: mit' in their front matter (Hub API cardData.license = mit for OpenGVLab/InternVL2-26B @46f3797 and OpenGVLab/InternVL2-4B @6eb21ec). The MIT grant covers 'this software and associated documentation files' and permits use, copy, modification, publication and distribution, including commercially, with the copyright and permission notice retained. Also within the owner's vendor-reported-score rule (each figure is OpenGVLab's own score for its own model).",
    "attribution": "OpenGVLab (Shanghai AI Laboratory), InternVL2-4B and InternVL2-26B model cards, MIT licence. Own-model Evaluation columns only; percentages divided by 100, OCRBench by 1000. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://huggingface.co/api/models/OpenGVLab/InternVL2-26B",
    "icon": "huggingface.co",
    "scope": "The InternVL2-4B and InternVL2-26B own-model columns of the Image and Video Benchmarks tables in the two pinned cards, pooled as one OpenGVLab evaluator (same team, toolkit and table layout). Competitor and earlier-InternVL columns excluded. MMBench test, CCBench, MMVet, MME, SEED, OpenCompass avg and MMBench-Video rows not used (no catalogue benchmark id or not the existing split)."
  },
  {
    "id": "llava-model-zoo-apache",
    "name": "LLaVA repository model zoo (haotian-liu/LLaVA docs/MODEL_ZOO.md)",
    "kind": "first-party",
    "match": [
      "github.com/haotian-liu/LLaVA/blob/c121f0432da27facab705978f83c4ada465e46fd/docs/MODEL_ZOO.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/haotian-liu/LLaVA/blob/c121f0432da27facab705978f83c4ada465e46fd/LICENSE",
    "terms_url": "https://github.com/haotian-liu/LLaVA/blob/c121f0432da27facab705978f83c4ada465e46fd/LICENSE",
    "quote": "Repository LICENSE at the pinned commit is Apache License 2.0 (GitHub API spdx_id Apache-2.0); s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce and distribute the Work, which includes documentation such as docs/MODEL_ZOO.md. The figures are also the LLaVA authors' own scores for their own checkpoints (vendor-reported rule).",
    "attribution": "Haotian Liu et al., LLaVA (haotian-liu/LLaVA) docs/MODEL_ZOO.md @c121f04, Apache-2.0. LLaVA-1.6 table, MMMU and TextVQA columns; divided by 100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://api.github.com/repos/haotian-liu/LLaVA/license",
    "icon": "github.com",
    "scope": "Own-checkpoint rows of the LLaVA-v1.6 table (MMMU val and TextVQA val columns) only. MathVista (split unstated), VQAv2 (test-dev), MME, MM-Bench and other columns not used; LLaVA-1.5 tables not used."
  },
  {
    "id": "first-party-zai-cogvlm2-scores",
    "name": "Zhipu AI (THUDM) CogVLM2 model card, Benchmark table (own models)",
    "kind": "first-party",
    "match": [
      "huggingface.co/zai-org/cogvlm2-llama3-chat-19B/blob/f592f291cf528389b2e4776b1e84ecdf6d71fbe3/README.md#benchmark"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/zai-org/cogvlm2-llama3-chat-19B/blob/f592f291cf528389b2e4776b1e84ecdf6d71fbe3/README.md",
    "quote": "Owner rule (vendor-reported scores, approved 2026-10-07): a vendor's own published score for its own model may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. The card's own licence is the custom 'cogvlm2' licence (license: other), so no open-licence claim is made.",
    "attribution": "Zhipu AI / THUDM, CogVLM2 model card (zai-org/cogvlm2-llama3-chat-19B @f592f29). Own-model rows only.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "huggingface.co",
    "scope": "The two own CogVLM2 rows (CogVLM2-LLaMA3, CogVLM2-LLaMA3-Chinese) of the card's Benchmark table; TextVQA, ChartQA and OCRBench columns only."
  },
  {
    "id": "hpsv3-readme-mit",
    "name": "HPSv3 repository README, image generation benchmark table (MizzenAI/HPSv3)",
    "kind": "leaderboard",
    "match": [
      "github.com/MizzenAI/HPSv3/blob/bd0c5fcb5f587617b0169c07222ab78d01e2f3c2/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://github.com/MizzenAI/HPSv3/blob/bd0c5fcb5f587617b0169c07222ab78d01e2f3c2/LICENSE",
    "terms_url": "https://github.com/MizzenAI/HPSv3/blob/bd0c5fcb5f587617b0169c07222ab78d01e2f3c2/LICENSE",
    "quote": "Repository licence MIT (GitHub API spdx_id MIT, root LICENSE at the pinned commit): 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies', with the copyright and permission notice retained. The README is repository documentation.",
    "attribution": "MizzenAI, HPSv3 (github.com/MizzenAI/HPSv3) README @bd0c5fc, MIT licence. Overall column of the image generation benchmark table. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://api.github.com/repos/MizzenAI/HPSv3",
    "icon": "github.com",
    "scope": "The 'Image Generation Benchmark of HPSv3' table, Overall column. Not used: 'SD2' (2.0 vs 2.1 not stated), 'Gemini 2.0 Flash' (experimental vs preview image endpoint not stated), Infinity, CogView4, PixArt-Sigma and HunyuanDiT (not catalogued or version not stated), per-category columns, and the paper (default arXiv licence)."
  },
  {
    "id": "paper-2512.01989v1-paibench",
    "name": "PAI-Bench: A Comprehensive Benchmark For Physical AI (arXiv:2512.01989v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2512.01989v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2512.01989v1",
    "quote": "arXiv abstract page for 2512.01989v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution. The HTML version prints 'License: CC BY 4.0'.",
    "attribution": "Fengzhe Zhou, Jiannan Huang, Jialuo Li, Deva Ramanan, Humphrey Shi (Georgia Tech, CMU), PAI-Bench, arXiv:2512.01989v1, CC BY 4.0. Table 3 Overall and Table 8 ELO Overall columns; percentages divided by 100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2512.01989v1",
    "icon": "arxiv.org",
    "scope": "Table 3 Overall and Table 8 human ELO Overall for the catalogued models the authors ran ('We evaluate all models using their default settings'). Generation settings are each model's defaults; CogVideoX/CogVideoX1.5 (T2V vs I2V checkpoint not stated), LTX-Video-2B (version not stated), Wan2.1-I2V-14B-720P, MAGI-1, DynamiCrafter and Cosmos-Predict rows not mapped."
  },
  {
    "id": "genexam-readme-mit",
    "name": "GenExam leaderboard (OpenGVLab/GenExam README, relaxed score)",
    "kind": "leaderboard",
    "match": [
      "github.com/OpenGVLab/GenExam/blob/1f820dbdb3fa802c2ea893acbf5ce5ada9d766d1/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://github.com/OpenGVLab/GenExam/blob/1f820dbdb3fa802c2ea893acbf5ce5ada9d766d1/LICENSE",
    "terms_url": "https://github.com/OpenGVLab/GenExam/blob/1f820dbdb3fa802c2ea893acbf5ce5ada9d766d1/LICENSE",
    "quote": "Repository licence MIT (GitHub API spdx_id MIT; root LICENSE at the pinned commit): permission to use, copy, modify, merge, publish, distribute, sublicense and/or sell copies of the software and associated documentation files, with the copyright and permission notice retained. The README leaderboard is repository documentation maintained by the GenExam authors.",
    "attribution": "OpenGVLab, GenExam (github.com/OpenGVLab/GenExam) README @1f820db, MIT licence. Relaxed Score Overall column; divided by 100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://api.github.com/repos/OpenGVLab/GenExam",
    "icon": "github.com",
    "scope": "Relaxed Score table, Overall column, for the 14 catalogued models that have no genexam-relaxed row from the arXiv paper entry visual-genexam. Not used: 'Qwen-Image-3.0' (non-Pro, not catalogued), 'Seedream 5.0' (Lite vs Pro not stated), 'BAGEL' non-thinking and 'Janus-Pro' (size not stated), '(w/o PE)' rows, and the rows that duplicate paper models. Owner may prefer to merge this entry with visual-genexam (same authors and judge)."
  },
  {
    "id": "paper-2503.12329v1-caparena",
    "name": "CapArena: Benchmarking and Analyzing Detailed Image Captioning in the LLM Era (arXiv:2503.12329v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2503.12329v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2503.12329v1",
    "quote": "arXiv abstract page for 2503.12329v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution. The HTML version prints 'License: CC BY 4.0'.",
    "attribution": "Kanzhi Cheng, Wenpo Song, Jiaxin Fan et al. (Nanjing University), CapArena, arXiv:2503.12329v1, CC BY 4.0. Table 8 Score_Avg column; model labels mapped to catalogue ids. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2503.12329v1",
    "icon": "arxiv.org",
    "scope": "Appendix J Table 8 (CapArena-Auto leaderboard) Score_Avg for 13 catalogued VLMs. Not used: Gemini-2.0-flash-exp, Ovis-2-34B, Ovis-1.6-27B, GLM-4V-Plus, InternVL2.5-8B, Qwen2-VL-2B, Cambrian-34B, LLaVA-1.5-7B (not catalogued) and Hunyuan-standard-vision (snapshot not stated); the human arena ranking (Figure 1, chart only) and per-baseline columns."
  },
  {
    "id": "paper-2501.16411v2-physbench",
    "name": "PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (arXiv:2501.16411v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2501.16411v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2501.16411v2",
    "quote": "arXiv abstract page for 2501.16411v2 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Wei Chow et al., PhysBench, arXiv:2501.16411v2, CC BY 4.0. Table 3 and Appendix Table 28 Avg column; divided by 100. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2501.16411v2",
    "icon": "arxiv.org",
    "scope": "Avg column of Table 3 and Appendix Table 28 for 10 catalogued VLMs whose checkpoint is unambiguous. Not used: GPT-4o, GPT-4V, Gemini-1.5-flash/pro, Claude-3.5-sonnet (snapshot not stated), MolmoE-7B-D/72B (naming does not match a released checkpoint), and uncatalogued models."
  },
  {
    "id": "paper-2505.20740v3-msearth",
    "name": "MSEarth: A Multimodal Benchmark for Earth Science Phenomenon Discovery with MLLMs (arXiv:2505.20740v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2505.20740v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2505.20740v3",
    "quote": "arXiv abstract page for 2505.20740v3 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution. The HTML version prints 'License: CC BY 4.0'.",
    "attribution": "Xiangyu Zhao et al., MSEarth, arXiv:2505.20740v3, CC BY 4.0. Table 4 Overall ACC column; divided by 100; model versions from Appendix Table 10. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2505.20740v3",
    "icon": "arxiv.org",
    "scope": "Table 4 Overall ACC for 6 catalogued models with stated or unique versions. Not used: GPT-4o (version truncated as 'GPT-4o-20'), GPT-5.2 and Gemini-3-Flash (not in the version table; reasoning setting unstated), Llama3.2-90B-Vision (base path), the authors' fine-tuned MSEarth models and uncatalogued models."
  },
  {
    "id": "paper-2505.17613v2-mmmg",
    "name": "MMMG: a Comprehensive and Reliable Benchmark for Multitask Multimodal Generation (arXiv:2505.17613v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2505.17613v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2505.17613v2",
    "quote": "arXiv abstract page for 2505.17613v2 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Jihan Yao, Yushi Hu, Wenyuan Wang, Bin Han et al., MMMG, arXiv:2505.17613v2, CC BY 4.0. Table 3 GenEval column and Appendix Table 9 image Average row; checkpoints from Appendix C. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2505.17613v2",
    "icon": "arxiv.org",
    "scope": "Table 3 GenEval column (the authors' own GenEval runs of nine image models) and Appendix Table 9 image-generation Average for the ten catalogued checkpoints listed in Appendix C. Not used: the Arena column (copied), DrawBench/GenAI-Bench VQAScore columns (GenAI-Bench is a 400-prompt subsample), Janus Pro and BLIP-3o (size unstated), interleaved, audio and speech results."
  },
  {
    "id": "paper-2506.10963v2-mmmg-kg",
    "name": "MMMG: A Massive, Multidisciplinary, Multi-Tier Generation Benchmark for Text-to-Image Reasoning (arXiv:2506.10963v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2506.10963v2"
    ],
    "exact": true,
    "license": "CC-BY-SA-4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://arxiv.org/abs/2506.10963v2",
    "quote": "arXiv abstract page for 2506.10963v2 links 'http://creativecommons.org/licenses/by-sa/4.0/' as the article licence (CC BY-SA 4.0). CC BY-SA 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material and Adapted Material, including commercially, subject to attribution and ShareAlike (s.3(b)) for adapted material.",
    "attribution": "Yuxuan Luo, Yuhui Yuan, Junwen Chen et al., MMMG (T2I reasoning), arXiv:2506.10963v2, CC BY-SA 4.0. Table 2 Avg column; divided by 100. Adapted values remain under CC BY-SA 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2506.10963v2",
    "icon": "arxiv.org",
    "scope": "Table 2 Avg for 7 catalogued models the authors evaluated. Not used: GPT-4o image generation (endpoint not stated), FLUX.1-[pro] (not catalogued), LlamaGen, JanusFlow, SimpleAR, CogView-4, SEED-X, Infinity, FLUX-Reason variants and the recaption/refiner rows."
  },
  {
    "id": "paper-2502.10663v1-real",
    "name": "REAL: Realism Evaluation of Text-to-Image Generation Models for Effective Data Augmentation (arXiv:2502.10663v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2502.10663v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2502.10663v1",
    "quote": "arXiv abstract page for 2502.10663v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (CC BY 4.0). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution. The HTML version prints 'License: CC BY 4.0'.",
    "attribution": "Ran Li, Xiaomeng Jin, Heng Ji, REAL, arXiv:2502.10663v1, CC BY 4.0. Table 5 Average Score column. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2502.10663v1",
    "icon": "arxiv.org",
    "scope": "Table 5 Average Score for DALL-E 3, Stable Diffusion 1.1 and Kandinsky 3. SD 3.5 Turbo not mapped."
  },
  {
    "id": "alpaca-eval-leaderboard-apache",
    "name": "AlpacaEval 1.0 and 2.0 leaderboards (tatsu-lab/alpaca_eval repository)",
    "kind": "leaderboard",
    "match": [
      "github.com/tatsu-lab/alpaca_eval/blob/cd543a149df89434d8a54582c0151c0b945c3d20/src/alpaca_eval/leaderboards/data_AlpacaEval_2/weighted_alpaca_eval_gpt4_turbo_leaderboard.csv",
      "github.com/tatsu-lab/alpaca_eval/blob/cd543a149df89434d8a54582c0151c0b945c3d20/src/alpaca_eval/leaderboards/data_AlpacaEval/alpaca_eval_gpt4_leaderboard.csv"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://github.com/tatsu-lab/alpaca_eval/blob/cd543a149df89434d8a54582c0151c0b945c3d20/LICENSE",
    "quote": "The repository LICENSE at the pinned commit is the Apache License, Version 2.0, and the leaderboard CSV is a committed file of that repository. Apache-2.0 section 2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce and distribute the Work and Derivative Works.",
    "attribution": "Dubois, Li, Taori et al., AlpacaEval (tatsu-lab/alpaca_eval), AlpacaEval 1.0 and 2.0 leaderboard CSVs @cd543a1, Apache-2.0. AlpacaEval 2.0 length-controlled and AlpacaEval 1.0 win rates only; dev (verbose/concise) entries and non-catalogue systems not used.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "github.com",
    "scope": "AlpacaEval 2.0 length-controlled and AlpacaEval 1.0 win rates for catalogue models whose exact checkpoint the entry's model config names."
  },
  {
    "id": "vectara-hallucination-leaderboard-apache",
    "name": "Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)",
    "kind": "leaderboard",
    "match": [
      "github.com/vectara/hallucination-leaderboard/blob/af0e3207741054b44e7c87688a40a815ecf446df/README.md",
      "github.com/vectara/hallucination-leaderboard/blob/9ceb7826f56f6cdafbbf92c85b3980a7718e66f0/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://github.com/vectara/hallucination-leaderboard/blob/af0e3207741054b44e7c87688a40a815ecf446df/LICENSE",
    "quote": "The repository LICENSE at the pinned commit is the Apache License, Version 2.0, and the leaderboard table is part of the committed README. Apache-2.0 section 2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce and distribute the Work and Derivative Works.",
    "attribution": "Vectara, Hallucination Leaderboard (vectara/hallucination-leaderboard @af0e320), Apache-2.0. Factual consistency rates for catalogue models whose exact model the README's API details identify. Also the previous-dataset table on branch hhem-2.3-old-dataset @9ceb782.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "github.com",
    "scope": "Factual consistency rates from the README table; rows whose API identity is ambiguous, or served as a third party's own quantization, are not used."
  },
  {
    "id": "tau2-bench-leaderboard-sierra",
    "name": "τ²/τ³-bench leaderboard: Sierra's own evaluation runs",
    "match": [
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-opus-4-5_sierra_2026-02-26/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-sonnet-4-5_sierra_2026-02-26/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gemini-3-flash_sierra_2026-03-02/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gemini-3-pro_sierra_2026-03-02/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/glm-5-think_sierra_2026-03-02/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gpt-5-2_sierra_2026-02-26/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/qwen3.5-397b-a17b-think_sierra_2026-03-02/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-fable-5_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-opus-4-6_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-opus-4-7_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-opus-4-8_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-opus-5_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gemini-2-5-pro_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gemini-3-1-pro-preview_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/glm-5-2_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gpt-5-4_sierra_2026-03-25/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gpt-5-5_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gpt-5-6-sol_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/grok-4-1-fast_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/grok-4-2_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/grok-4-5_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/grok-4-fast_sierra_2026-05-05/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/inkling_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/kimi-k3_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/muse-spark-1-1_sierra_2026-08-04/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-3-7-sonnet_anthropic_2024-06-20/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gpt-4-1_openai_2024-06-20/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gpt-4-1-mini_openai_2024-06-20/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/gpt-5_sierra_2025-08-09/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/o4-mini_openai_2024-06-20/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/qwen3-8-max_sierra_2026-08-04/submission.json"
    ],
    "kind": "leaderboard",
    "exact": true,
    "license": "MIT",
    "license_url": "https://opensource.org/license/mit",
    "terms_url": "https://github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/LICENSE",
    "quote": "The repository LICENSE at the pinned commit is the MIT License (Copyright (c) 2025 Sierra Research): 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute ...'. The leaderboard submission files are committed files of the repository.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "github.com",
    "attribution": "Sierra Research, τ-bench leaderboard submissions run by Sierra (sierra-research/tau2-bench @4ce7c03), MIT. pass^1 per domain.",
    "scope": "pass^1 per domain from the committed submission.json files listed in match; each submitting organisation is a separate evaluator."
  },
  {
    "id": "tau2-bench-leaderboard-anthropic",
    "name": "τ-bench leaderboard: Anthropic's own submissions",
    "match": [
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-opus-4-1_anthropic_2025-01-15/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-opus-4_anthropic_2025-05-22/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-sonnet-4_anthropic_2025-05-22/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/claude-sonnet-4-5_anthropic_2025-10-02/submission.json"
    ],
    "kind": "leaderboard",
    "exact": true,
    "license": "MIT",
    "license_url": "https://opensource.org/license/mit",
    "terms_url": "https://github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/LICENSE",
    "quote": "The repository LICENSE at the pinned commit is the MIT License (Copyright (c) 2025 Sierra Research): 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute ...'. The leaderboard submission files are committed files of the repository.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "github.com",
    "attribution": "Anthropic's own τ-bench submissions in the sierra-research/tau2-bench leaderboard (@4ce7c03), MIT. Vendor-run; prompt addendum noted per row.",
    "scope": "pass^1 per domain from the committed submission.json files listed in match; each submitting organisation is a separate evaluator."
  },
  {
    "id": "tau2-bench-leaderboard-moonshot",
    "name": "τ²-bench leaderboard: Moonshot AI's own submission",
    "match": [
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/kimi-k2_moonshot-ai_2025-07-11/submission.json"
    ],
    "kind": "leaderboard",
    "exact": true,
    "license": "MIT",
    "license_url": "https://opensource.org/license/mit",
    "terms_url": "https://github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/LICENSE",
    "quote": "The repository LICENSE at the pinned commit is the MIT License (Copyright (c) 2025 Sierra Research): 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute ...'. The leaderboard submission files are committed files of the repository.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "github.com",
    "attribution": "Moonshot AI's own τ²-bench submission in the sierra-research/tau2-bench leaderboard (@4ce7c03), MIT. Vendor-run.",
    "scope": "pass^1 per domain from the committed submission.json files listed in match; each submitting organisation is a separate evaluator."
  },
  {
    "id": "charxiv-readme-apache",
    "name": "CharXiv leaderboard (princeton-nlp/CharXiv README)",
    "kind": "leaderboard",
    "match": [
      "github.com/princeton-nlp/CharXiv/blob/7ebe88f78dee387691551f071abcb2b9e1a8025b/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://github.com/princeton-nlp/CharXiv/blob/7ebe88f78dee387691551f071abcb2b9e1a8025b/LICENSE",
    "quote": "The repository is licensed under the Apache License, Version 2.0 (GitHub licence metadata and LICENSE file at the pinned commit), and the results table is part of the committed README. The underlying dataset is separately CC BY-SA 4.0. The README states these are the authors' own evaluations: 'We release full results on the validation set ... for all models we tested'.",
    "attribution": "Wang et al., CharXiv (princeton-nlp/CharXiv @7ebe88f), Apache-2.0. Reasoning and descriptive overall accuracy from the README table.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "github.com",
    "scope": "Validation-set overall accuracies from the README table, for catalogue models whose snapshot the paper or the release timing pins."
  },
  {
    "id": "paper-2411.19146v5-puzzle",
    "name": "Puzzle: Distillation-Based NAS for Inference-Optimized LLMs (arXiv:2411.19146v5)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2411.19146v5"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2411.19146v5",
    "quote": "arXiv abstract page for 2411.19146v5 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; the HTML version prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including commercially, subject to attribution.",
    "attribution": "Akhiad Bercovich et al., Puzzle: Distillation-Based NAS for Inference-Optimized LLMs, arXiv:2411.19146v5, CC BY 4.0. Table 2 non-chat rows. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2411.19146v5",
    "icon": "arxiv.org",
    "scope": "Table 2 Winogrande, ARC Challenge, MMLU, HellaSwag and GSM8K rows for Llama-3.1-Nemotron-51B-Instruct and Llama-3.1-70B-Instruct. Not used: TruthfulQA (variant not stated), XLSum, chat-prompt rows, Instruct HumanEval (CodeParrot variant), MT-Bench (judge not stated), RULER subset (partial subset of tasks), Nemotron-49B-Base tables (base model not catalogued)."
  },
  {
    "id": "first-party-prime-rl-scores",
    "name": "PRIME-RL Eurus-2-7B-PRIME model card, Evaluation table (own models)",
    "kind": "first-party",
    "match": [
      "huggingface.co/PRIME-RL/Eurus-2-7B-PRIME/blob/2f47d199012a5d37d0dcedbeff1e2802a0ea443c/README.md#evaluation"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/PRIME-RL/Eurus-2-7B-PRIME/blob/2f47d199012a5d37d0dcedbeff1e2802a0ea443c/README.md",
    "quote": "Owner rule (vendor-reported scores, approved 2026-10-07): a vendor's own published score for its own model may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. The card's own licence is apache-2.0.",
    "attribution": "PRIME-RL (Tsinghua/UIUC et al.), Eurus-2-7B-PRIME model card @2f47d19. Own-model rows only.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "huggingface.co",
    "scope": "Eurus-2-7B-PRIME and Eurus-2-7B-SFT columns for AIME 2024, MATH-500 and Minerva Math. Not used: AMC and OlympiadBench (no benchmark id), Qwen/Llama/GPT-4o columns (other labs)."
  },
  {
    "id": "paper-2405.18750v2-t2vturbo",
    "name": "T2V-Turbo: Breaking the Quality Bottleneck of Video Consistency Model with Mixed Reward Feedback (arXiv:2405.18750v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2405.18750v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2405.18750v2",
    "quote": "arXiv abstract page for 2405.18750v2 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Jiachen Li, Weixi Feng, Tsu-Jui Fu, Xinyi Wang et al., T2V-Turbo, arXiv:2405.18750v2, CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2405.18750v2",
    "icon": "arxiv.org",
    "scope": "Table 1 VBench Total for T2V-Turbo (VC2) and T2V-Turbo (MS) only (authors' own models). Not used: baseline rows (copied leaderboard values), sub-dimension scores."
  },
  {
    "id": "paper-2604.25072v1-xtcbench",
    "name": "Beyond Accuracy: Benchmarking Cross-Task Consistency in Unified Multimodal Models (XTC-Bench, arXiv:2604.25072v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2604.25072v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2604.25072v1",
    "quote": "arXiv abstract page for 2604.25072v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Authors of arXiv:2604.25072 (XTC-Bench), CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2604.25072v1",
    "icon": "arxiv.org",
    "scope": "Table 2 'Overall Gen.' for BAGEL-7B, BLIP3-o-8B, Janus-Pro-7B, OmniGen-2 and Show-o2-7B. Not used: Show-o (checkpoint unstated), MMaDA-8B and Tar-7B (not catalogued), GPT/Gemini pipeline rows, understanding and CCTA columns."
  },
  {
    "id": "first-party-showlab-show-o2-paper",
    "name": "Show Lab Show-o2 paper (arXiv:2506.15564v3), GenEval and DPG-Bench tables (own models)",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2506.15564v3"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2506.15564v3",
    "quote": "Owner rule (vendor-reported scores, approved 2026-10-07): a vendor's own published score for its own model may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. The document is a maker-authored technical report (default arXiv licence), recorded under the vendor-reported rule as maker documentation; see report for the open decision.",
    "attribution": "Show Lab, National University of Singapore, Show-o2: Improved Native Unified Multimodal Models, arXiv:2506.15564v3. Own-model figures only.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "arxiv.org",
    "scope": "Table 5 GenEval Overall and Table 6 DPG-Bench Overall for Show-o2 1.5B and 7B only. The paper itself carries arXiv's default non-exclusive licence (no open licence claimed); only the makers' own figures for their own models are recorded by hand. Not used: all other methods' rows, Show-o (checkpoint/resolution unstated), OneIG table (no overall printed), VBench tables."
  },
  {
    "id": "paper-2306.02561v3-llmblender",
    "name": "LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusion (arXiv:2306.02561v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2306.02561v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2306.02561v3",
    "quote": "arXiv abstract page for 2306.02561v3 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Dongfu Jiang, Xiang Ren, Bill Yuchen Lin, LLM-Blender, arXiv:2306.02561v3, CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2306.02561v3",
    "icon": "arxiv.org",
    "scope": "Table 2 GPT-Rank column for the five LLMs that match catalogue checkpoints (Open Assistant SFT-4 Pythia 12B, ChatGLM-6B, Dolly V2 12B, StableLM-Tuned-Alpha 7B, Flan-T5-XXL). Not used: other candidates, oracle/ranker/ensemble rows, BERTScore/BARTScore/BLEURT/threshold columns."
  },
  {
    "id": "paper-2508.13680v5-vmmu",
    "name": "VMMU: A Vietnamese Multitask Multimodal Understanding and Reasoning Benchmark (arXiv:2508.13680v5)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2508.13680v5"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2508.13680v5",
    "quote": "arXiv abstract page for 2508.13680v5 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Authors of arXiv:2508.13680 (VMMU), CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2508.13680v5",
    "icon": "arxiv.org",
    "scope": "Table 2 Mean for eight open-weight VLMs that match catalogue checkpoints. Not used: closed-source rows (snapshots unstated), Aya-Vision-8B, Mistral-Medium-3 and other uncatalogued models, per-task columns, OCR-ablation tables."
  },
  {
    "id": "paper-2605.01417v1-medmarks",
    "name": "Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks (arXiv:2605.01417v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.01417v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.01417v1",
    "quote": "arXiv abstract page for 2605.01417v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Authors of arXiv:2605.01417 (Medmarks), CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2605.01417v1",
    "icon": "arxiv.org",
    "scope": "Table 4 (all models by mean win rate on Medmarks-V) for 30 configurations that match catalogue models by exact name. Not used: quantised variants, non-default reasoning efforts, rows with ambiguous revisions (Qwen3 Thinking/Instruct without 2507, Magistral Small), uncatalogued models, Medmarks-OE and per-benchmark scores."
  },
  {
    "id": "first-party-alibaba-wan-paper",
    "name": "Alibaba Wan technical report (arXiv:2503.20314v2), VBench table (own models)",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2503.20314v2"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2503.20314v2",
    "quote": "Owner rule (vendor-reported scores, approved 2026-10-07): a vendor's own published score for its own model may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. The document is a maker-authored technical report (default arXiv licence), recorded under the vendor-reported rule as maker documentation; see report for the open decision.",
    "attribution": "Alibaba Wan team, Wan: Open and Advanced Large-Scale Video Generative Models, arXiv:2503.20314v2. Own-model figures only.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "icon": "arxiv.org",
    "scope": "VBench Total Score for Wan 1.3B and Wan 14B (2025-02-24) only. The report carries arXiv's default non-exclusive licence (no open licence claimed); only the makers' own figures for their own models are recorded by hand. Not used: competitor rows, Wan-Bench (no catalogue benchmark id; competitors anonymised)."
  },
  {
    "id": "paper-2501.03575v3-cosmos",
    "name": "Cosmos World Foundation Model Platform for Physical AI (arXiv:2501.03575v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2501.03575v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2501.03575v3",
    "quote": "arXiv abstract page for 2501.03575v3 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "NVIDIA, Cosmos World Foundation Model Platform for Physical AI, arXiv:2501.03575v3, CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2501.03575v3",
    "icon": "arxiv.org",
    "scope": "Table 19 rows for Cosmos-Predict1-7B-Text2World, 7B-Video2World, 4B and 5B-Video2World (Sampson error and pose-estimation success rate). Not used: VideoLDM row, real-video reference, PSNR/SSIM/LPIPS columns, post-trained variants, physics-alignment table (Video2World only)."
  },
  {
    "id": "paper-2504.07072v2-kaleidoscope",
    "name": "Kaleidoscope: In-language Exams for Massively Multilingual Vision Evaluation (arXiv:2504.07072v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2504.07072v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2504.07072v2",
    "quote": "arXiv abstract page for 2504.07072v2 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Israfel Salazar et al., Kaleidoscope, arXiv:2504.07072v2, CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2504.07072v2",
    "icon": "arxiv.org",
    "scope": "Overall Acc. column for six models matching catalogue checkpoints (Aya-Vision-32B, Qwen2.5-VL-72B/32B, Molmo-7B-D, GPT-4o 2024-08-06, Claude 3.5 Sonnet 2024-10-22). Not used: Gemini 1.5 Pro (snapshot unstated), other uncatalogued models, multimodal/text-only/valid-response columns."
  },
  {
    "id": "paper-2404.16164v1-factrecall",
    "name": "Towards a Holistic Evaluation of LLMs on Factual Knowledge Recall (arXiv:2404.16164v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2404.16164v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2404.16164v1",
    "quote": "arXiv abstract page for 2404.16164v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Authors of arXiv:2404.16164, CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2404.16164v1",
    "icon": "arxiv.org",
    "scope": "Table 1 EM columns (zero-shot, 10-shot) for fourteen open models matching catalogue checkpoints. Not used: closed GPT rows, uncatalogued variants, 'Contains' columns, 1-/6-shot columns, full 15k results (Table 4), counterfactual experiments."
  },
  {
    "id": "paper-2503.16167v2-codereviewqa",
    "name": "CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2503.16167v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2503.16167v2",
    "quote": "arXiv abstract page for 2503.16167v2 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Authors of arXiv:2503.16167 (CodeReviewQA), CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2503.16167v2",
    "icon": "arxiv.org",
    "scope": "Overall columns of the ACR exact-match table (Table 8) and the CTR, CL-hard and SI-hard invariant-accuracy tables (Tables 9, 11, 13) for the 48 evaluated models whose Hugging Face repo equals a catalogue repo. Not used: easy CL/SI variants, per-language columns, PPA, advanced-prompting experiments, unmatched models."
  },
  {
    "id": "paper-2505.22113v1-thinkbench",
    "name": "THINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Models (arXiv:2505.22113v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2505.22113v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2505.22113v1",
    "quote": "arXiv abstract page for 2505.22113v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Authors of arXiv:2505.22113 (THINK-Bench), CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2505.22113v1",
    "icon": "arxiv.org",
    "scope": "Table 2 Accuracy and Efficiency columns for eight reasoning models matching catalogue entries. Not used: DeepSeek-R1 (revision unstated), uncatalogued models, recall/precision/reflection/token columns."
  },
  {
    "id": "paper-2410.14897v1-gencopa",
    "name": "From Test-Taking to Test-Making: Examining LLM Authoring of Commonsense Assessment Items (arXiv:2410.14897v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2410.14897v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2410.14897v1",
    "quote": "arXiv abstract page for 2410.14897v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Melissa Roemmele and Andrew S. Gordon, arXiv:2410.14897v1, CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-08",
    "evidence_url": "https://arxiv.org/abs/2410.14897v1",
    "icon": "arxiv.org",
    "scope": "Table 3 COPA test accuracy for nine models matching catalogue checkpoints. Not used: Mistral-7B-v0.3 and MPT-30B (not catalogued), Gen-COPA tables."
  },
  {
    "id": "paper-2510.01141v1-apriel-1-5",
    "name": "Apriel-1.5-15b-Thinker (arXiv:2510.01141v1), Table 4 multimodal benchmarks",
    "kind": "paper",
    "match": [
      "arxiv.org/abs/2510.01141v1",
      "arxiv.org/html/2510.01141v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2510.01141v1",
    "quote": "arXiv abstract page for 2510.01141v1 (the only version, submitted 1 Oct 2025) links 'http://creativecommons.org/licenses/by/4.0/' as the article licence. CC BY 4.0 s.2(a)(1) permits reproduction and sharing, including commercially, with attribution. §5.2: 'We evaluate vision capabilities using the VLMEvalKit toolkit ... For each dataset, we adhere to official or community-standard protocols as implemented in VLMEvalKit and adopt consistent prompts and inference settings across models to ensure fair comparisons.'",
    "attribution": "Shruthan Radhakrishna et al. (ServiceNow SLAM Lab), Apriel-1.5-15b-Thinker, arXiv:2510.01141v1, CC BY 4.0. Table 4 CharXiv descriptive/reasoning cells. No endorsement.",
    "scope": "Table 4 CharXiv 'Des' and 'R' columns for Llama 4 Maverick, Qwen 2.5 VL 32B-Instruct and Magistral Small-2509 only (the other catalogue models in Table 4 carry identical ServiceNow numbers in the Apriel-1.6 card and are recorded from there). Not used: Table 3 (credited to Artificial Analysis), GPT-5 Minimal column (lower setting), Kimi-VL-2506 and Apriel-1.5 (not catalogued).",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "inspect-charxiv-meridianlabs",
    "name": "Meridian Labs inspect_charxiv evaluation report (GitHub README)",
    "kind": "leaderboard",
    "match": [
      "github.com/meridianlabs-ai/inspect_charxiv/blob/317e4ff4335d7d4242d8e3c803898b7345973bd2/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://github.com/meridianlabs-ai/inspect_charxiv/blob/317e4ff4335d7d4242d8e3c803898b7345973bd2/LICENSE",
    "terms_url": "https://github.com/meridianlabs-ai/inspect_charxiv/blob/317e4ff4335d7d4242d8e3c803898b7345973bd2/LICENSE",
    "quote": "LICENSE at the pinned commit: 'MIT License / Copyright (c) 2026 Meridian Labs / Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute ...'; GitHub reports MIT. README 'License and attribution': 'Code in this repository is MIT-licensed. Question/answer content from the CharXiv dataset ... is CC BY-SA 4.0'. The Evaluation Report table is part of the committed README (documentation). The UK AISI inspect_evals register (register/charxiv/eval.yaml) pins this exact commit.",
    "attribution": "Meridian Labs (contributor jchengjr), inspect_charxiv @317e4ff, MIT. Evaluation Report accuracies. No endorsement.",
    "scope": "Evaluation Report rows for claude-fable-5, gpt-4o (alias, = gpt-4o-2024-08-06) and gpt-6-astra, descriptive (4,000) and reasoning (1,000) validation subsets.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-qwen-scores",
    "name": "Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card)",
    "kind": "first-party",
    "match": [
      "qwen.ai/blog",
      "arxiv.org/html/2502.13923v1",
      "arxiv.org/pdf/2511.21631v2",
      "huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md",
      "arxiv.org/html/2409.12186v3",
      "arxiv.org/html/2409.12191v2",
      "arxiv.org/html/2409.12122v1"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://qwen.ai/blog?id=qwen3.8",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model, from its announcement, model card, tech report or docs, may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'; removed if the vendor objects. The qwen.ai posts show no licence or terms statement; arXiv 2502.13923v1 and 2511.21631v2 carry arXiv's non-exclusive distribution licence (http://arxiv.org/licenses/nonexclusive-distrib/1.0/, checked on the abs pages); the Qwen3.8-Flash-Next card is under the 'Qwen Community License 1.0' (Model-as-a-Service / AI Work Assistant commercial restriction). No open-licence claim is made.",
    "attribution": "Qwen Team (Alibaba), vendor-reported results for Qwen's own models from qwen.ai launch posts, the Qwen2.5-VL and Qwen3-VL technical reports and the pinned Qwen3.8-Flash-Next card; individual figures hand-recorded 2026-10-09; percentages converted to fractions. Lab-reported, not independent. No endorsement.",
    "scope": "CharXiv (RQ) without code interpreter and CharXiv (DQ) cells for Qwen's own models only. With-CI cells, other labs' columns, and figures that merely repeat a value already counted in another Qwen card run are not used. The matcher ignores query strings, so 'qwen.ai/blog' covers every qwen.ai/blog?id=... post (all Qwen own posts). qwen.ai/blog is client-rendered; content is served by https://qwen.ai/api/v2/article/?language=en-US&path=<id>&type=qwen_ai. Extended 2026-10-09 (pass 8, group E): Qwen3-VL Technical Report arXiv 2511.21631v2 (already in match as arxiv.org/pdf/2511.21631v2) — Qwen's own Qwen3-VL and Qwen3 columns of Tables 2-4 (vision: MMMU, MMMU-Pro, MathVista mini, RealWorldQA, MMStar, HallusionBench, DocVQA test, ChartQA test, OCRBench, CC-OCR, CharXiv DQ/RQ, MVBench, Video-MME w/o sub, LVBench, ScreenSpot Pro, AndroidWorld) and Tables 5-10 (text: MMLU-Pro, MMLU-Redux, GPQA, SuperGPQA, AIME-25, LiveCodeBench v6, IFEval, BFCL-v3, TAU2-Telecom), hand-recorded one figure at a time; hybrid Qwen3 checkpoints only in their thinking columns; figures already counted in another Qwen card run are skipped; Gemini/GPT/Claude/DeepSeek/o3 columns not used. No new URL prefix needed.",
    "icon": "qwenlm.github.io",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-moonshot-scores",
    "name": "Moonshot AI vendor-reported benchmark scores for its own Kimi models (Hugging Face model cards)",
    "kind": "first-party",
    "match": [
      "huggingface.co/moonshotai/Kimi-K2.5/blob/4d01dfe0332d63057c186e0b262165819efb6611/README.md",
      "huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md",
      "huggingface.co/moonshotai/Kimi-K3/blob/f831ab66814297da540d832a5235f8e904f29d06/README.md"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/moonshotai/Kimi-K3/blob/f831ab66814297da540d832a5235f8e904f29d06/LICENSE",
    "quote": "Owner rule (b), approved 2026-10-07 (vendor's own score for its own model, hand-recorded one figure at a time; other labs' cells excluded; by: 'lab'). Licences at the pinned revisions: Kimi-K2.5 and Kimi-K2.6 LICENSE = 'Modified MIT License' - the standard MIT grant ('Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ...') plus 'Our only modification part is that, if the Software (or any derivative works thereof) is used for any of your commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars (or equivalent in other currencies) in monthly revenue, you shall prominently display \"Kimi K2.5\" [resp. \"Kimi K2.6\"] on the user interface of such product or service.' Kimi-K3 LICENSE = 'Kimi K3 License': MIT-style grant but a Model-as-a-Service business with >US$20M revenue over 12 months 'must enter into a separate agreement with Moonshot AI before using the Software or its derivative works for any commercial purpose', plus the same UI-display clause. Because K3 is not open and the K2.x licence is a modified MIT, the entry relies on rule (b) rather than an open-licence claim.",
    "attribution": "Moonshot AI, vendor-reported results for its own Kimi K2.5, K2.6 and K3 models from the pinned Hugging Face model cards; hand-recorded 2026-10-09; percentages converted to fractions. Lab-reported, not independent. No endorsement.",
    "scope": "Own Kimi columns only, CharXiv (RQ) without tools. 'w/ python' / with-tool cells and all other labs' columns (several are re-evaluations marked * or cited from Artificial Analysis / Vals AI) are not used. Alternative: if the owner treats Modified MIT as open, K2.5/K2.6 could get card2-moonshotai-kimi-k2-5 / card2-moonshotai-kimi-k2-6 entries instead, but then each card is a one-model CharXiv run (K2.6 card + its K2.5 column would be two).",
    "icon": "huggingface.co",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "first-party-xiaomi-scores",
    "name": "Xiaomi MiMo vendor-reported benchmark scores for its own models (mimo.xiaomi.com launch pages)",
    "kind": "first-party",
    "match": [
      "mimo.xiaomi.com/mimo-v2-omni",
      "mimo.xiaomi.com/mimo-v2-5",
      "mimo.xiaomi.com/mimo-v2-pro"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://mimo.xiaomi.com/",
    "quote": "Owner rule (b), approved 2026-10-07 (vendor's own score for its own model, hand-recorded one figure at a time; other labs' cells excluded; by: 'lab'). The launch pages carry 'Xiaomi MiMo Team · 2026'; the mimo.xiaomi.com home page footer reads '© 2010 - 2026 Xiaomi. All Rights Reserved'. No licence is granted, so no open-licence claim is made. Kept separate from first-party-xiaomi, which covers only specifications.",
    "attribution": "Xiaomi MiMo Team, vendor-reported results for its own MiMo models from the mimo.xiaomi.com launch pages (MiMo-V2-Omni, 2026-03-18; MiMo-V2.5, 2026-04-22); hand-recorded 2026-10-09; percentages converted to fractions. Lab-reported. No endorsement.",
    "scope": "CharXiv RQ values for MiMo's own models, from the chart data in the page HTML that the page renders as text labels. Other labs' values on the same charts are not used.",
    "verdict": "allowed",
    "checked": "2026-10-11"
  },
  {
    "id": "card2-coherelabs-north-micro-vision-instruct",
    "name": "CohereLabs/North-Micro-Vision-Instruct model card (benchmark table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/CohereLabs/North-Micro-Vision-Instruct/blob/fa548d9aa62c1220bfd21fc6cfc465f1bd2a1a40/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/CohereLabs/North-Micro-Vision-Instruct/blob/fa548d9aa62c1220bfd21fc6cfc465f1bd2a1a40/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0' (HF API cardData.license 'apache-2.0', repo not gated); card text: 'North Micro Vision Instruct is a 2.4B-parameter open-weight vision-language model with native-resolution image support, released under the Apache 2.0 license.' No separate LICENSE file in the repo (files: README.md, config/tokenizer/preprocessor JSONs, chat_template.jinja, model.safetensors). Benchmark text: 'We ran vision-language and text-only evaluations with VLMEvalKit, capping generation at 1,024 tokens'.",
    "attribution": "Cohere (Team Cohere / Cohere Labs). CohereLabs/North-Micro-Vision-Instruct model card, revision fa548d9aa62c1220bfd21fc6cfc465f1bd2a1a40. Source: https://huggingface.co/CohereLabs/North-Micro-Vision-Instruct/blob/fa548d9aa62c1220bfd21fc6cfc465f1bd2a1a40/README.md. Licence: Apache-2.0. Changes: selected CharXiv (DQ) cells transcribed into JSON; benchmark names clarified. No endorsement.",
    "scope": "Benchmark Results table, row CharXiv_DQ only (this pass): Cohere's own VLMEvalKit runs of North-Micro-Vision-Instruct and seven comparison models (Ministral-3-3B-Instruct, LFM2.5-VL-1.6B, Phi-3.5-vision-instruct, Gemma-4-E2B-it, Qwen3-VL-2B-Instruct, Qwen3.5-2B-Instruct, SmolVLM2.2B). Linked HF blog post used only for methodology wording, not numbers. Extended 2026-10-09 (pass 8, group E): also rows MMBench_DEV_EN_V11, MMStar, RealWorldQA, ChartQA_Test, OCRBench, AI2D_TEST, CountBench, HallusionBench, MMLU_test, MMLU-Pro_test, Multi-If and IFEval for the catalogued comparison columns (Ministral-3-3B-Instruct, Phi-3.5-vision-instruct, Gemma-4-E2B-it, Qwen3.5-2B-Instruct) and the proposed qwen3-vl-2b-instruct; same VLMEvalKit statement. Rows not mapped: GQA, MTL-MMBench, MMMB, BLINK, DocVQA_VAL/InfoVQA_VAL (val split; docvqa id is test), OCRBench_v2, MMMU_DEV_VAL (dev+val, not the val-only mmmu id), RefCOCO.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "paper-2607.20062v2-solar-open-2",
    "name": "Solar Open 2 Technical Report (Upstage, arXiv:2607.20062v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2607.20062v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2607.20062v2",
    "quote": "arXiv abstract pages for 2607.20062v1 and 2607.20062v2 both link 'http://creativecommons.org/licenses/by/4.0/' as the article licence; the HTML prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including commercially, subject to attribution.",
    "attribution": "Sungrae Park et al. (Upstage), Solar Open 2 Technical Report, arXiv:2607.20062v2, CC BY 4.0. Table 3 cells for non-Upstage models; percentages converted to fractions. No endorsement.",
    "scope": "Table 3 (English benchmarks, 'All numbers are measured with our internal evaluation harness'): Command A+, Mistral Medium 3.5 (high), MiMo-V2.5 and DeepSeek-V4-Flash (max; April preview weights) on MMLU-Pro, GPQA-Diamond, HLE (no tools), LiveCodeBench v6, HMMT Feb 2026, AIME 2026, Multi-Challenge, IFBench, SWE-Bench Verified, Terminal Bench Hard, APEX-Agents, MCP-Atlas and the tau3 banking domain. Not used: Solar Open 2 / Solar Open 1 columns (not catalogued), ArtifactsBench and AA-LCR (no benchmark id), GDPval-AA v2 ELO (an Artificial Analysis leaderboard Elo 'measured on 30 June 2026', held), Table 4 Korean benchmarks (out of this pass's scope).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "evidence_url": "https://arxiv.org/abs/2607.20062v2"
  },
  {
    "id": "paper-2601.03286v1-hyperclova-x-32b-think",
    "name": "HyperCLOVA X 32B Think (NAVER Cloud, arXiv:2601.03286v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2601.03286v1",
      "arxiv.org/abs/2601.03286v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2601.03286v1",
    "quote": "arXiv abstract page for 2601.03286v1 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence (only version). CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including commercially, subject to attribution.",
    "attribution": "NAVER Cloud HyperCLOVA X Team, HyperCLOVA X 32B Think, arXiv:2601.03286v1, CC BY 4.0. Table 5 Terminal Bench Hard cells; percentages converted to fractions. No endorsement.",
    "scope": "Table 5 row 'Terminal Bench Hard' (47-task Artificial Analysis split, Terminus-2, reasoning enabled, authors' own Omni-Evaluator runs): EXAONE 4.0 32B and Qwen3-VL 32B-Thinking (comparison-only model proposed); HyperCLOVA X 32B Think row held (model not catalogued). Other Table 5/6 rows not reviewed in this pass (Tau2 rows noted as a lead).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "evidence_url": "https://arxiv.org/abs/2601.03286v1"
  },
  {
    "id": "multiverse-hypernova-60b-2602-card",
    "name": "Multiverse Computing HyperNova 60B 2602 model card: evaluation table (own runs incl. gpt-oss baselines)",
    "kind": "dataset",
    "match": [
      "huggingface.co/MultiverseComputingCAI/Hypernova-60B-2602/blob/382391945c26d1ab65fc8bdc174865067cf80cd1/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/MultiverseComputingCAI/Hypernova-60B-2602/blob/382391945c26d1ab65fc8bdc174865067cf80cd1/README.md",
    "quote": "Card front matter 'license: apache-2.0' and the 'License' badge (Apache 2.0) at @3823919. Apache-2.0 s.2 grants a perpetual, worldwide, royalty-free copyright licence to reproduce and distribute the Work, including commercially, with notice. The card states 'Reported numbers use the methodology described above' (Multiverse's own runs).",
    "attribution": "Multiverse Computing, HyperNova 60B 2602 model card (MultiverseComputingCAI/Hypernova-60B-2602 @3823919, Apache-2.0), Terminal Bench (TB-Hard) row; integers converted to fractions. No endorsement.",
    "scope": "Row 'Terminal Bench' (= Terminal-Bench Hard, AA subset, harbor 0.1.43, terminus-2, reasoning high, 3 repeats): gpt-oss-20b 9 and gpt-oss-120b 22. HyperNova 60B 2602 (16) not catalogued. Other rows (MMLU-Pro, BFCL v4, Tau2-bench Telecom, AIME25, GPQA:d, IFBench, SciCode, LiveCodeBench, AA-LCR, AA-Omniscience) not reviewed in this pass; Tau2 Telecom gpt-oss-20b 59 / gpt-oss-120b 68 is a lead for the tau2 topic.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "nvidia-modelopt-card-baselines",
    "name": "NVIDIA ModelOpt quantized-checkpoint cards: baseline (unquantized) measurements of third-party models",
    "kind": "dataset",
    "match": [
      "huggingface.co/nvidia/Gemma-4-31B-IT-NVFP4/blob/4135a98a9b728a548947683219633b25682223ac/README.md",
      "huggingface.co/nvidia/DeepSeek-V4-Pro-0813-NVFP4/blob/949138637e8e8fe190335be2396af62187dc8a83/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0 (Gemma-4-31B-IT-NVFP4 repo) / MIT (DeepSeek-V4-Pro-0813-NVFP4 repo)",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/nvidia/DeepSeek-V4-Pro-0813-NVFP4/blob/949138637e8e8fe190335be2396af62187dc8a83/README.md",
    "quote": "nvidia/Gemma-4-31B-IT-NVFP4 @4135a98 front matter 'license_name: apache-license-2.0, license_link: https://ai.google.dev/gemma/apache_2' and section 'License and Terms of Use: Apache License 2.0'; nvidia/DeepSeek-V4-Pro-0813-NVFP4 @9491386 front matter 'license: mit' and 'License/Terms of Use: [MIT]'. Both licences permit commercial reuse with notice. The baseline rows are NVIDIA's own measurements of the unquantized source checkpoints.",
    "attribution": "NVIDIA ModelOpt, evaluation tables of nvidia/Gemma-4-31B-IT-NVFP4 (@4135a98, Apache-2.0) and nvidia/DeepSeek-V4-Pro-0813-NVFP4 (@9491386, MIT); baseline rows only; percentages converted to fractions. No endorsement.",
    "scope": "Terminal-Bench Hard baseline rows only: Gemma 4 31B IT BF16 27.08 and DeepSeek-V4-Pro-0813 (MXFP4 source) 51.39. Quantized-checkpoint rows not used (they belong in quant.json if wanted). Pooling the two cards into one NVIDIA ModelOpt evaluator run is a judgement call (same team, different sampling settings); see held_for_owner.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "tau2-bench-leaderboard-qwen",
    "name": "τ²-bench leaderboard: Qwen (Alibaba) own submissions",
    "kind": "leaderboard",
    "match": [
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/qwen3-max_qwen_2024_09_23/submission.json",
      "github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/web/leaderboard/public/submissions/qwen3-max_qwen_2026-01-23/submission.json"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://opensource.org/license/mit",
    "terms_url": "https://github.com/sierra-research/tau2-bench/blob/4ce7c0397c1eb65c9bbe59aeacfe1ca44a1cd699/LICENSE",
    "quote": "The repository LICENSE at the pinned commit is the MIT License (Copyright (c) 2025 Sierra Research): 'Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files ... to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute ...'. The leaderboard submission files are committed files of the repository (same terms as the existing tau2-bench-leaderboard-* entries).",
    "attribution": "Qwen team (Alibaba)'s own τ²-bench submissions in the sierra-research/tau2-bench leaderboard (@4ce7c03), MIT. Vendor-run / vendor-reported.",
    "scope": "Telecom pass^1 (also airline/retail if wanted) from the two committed Qwen submission.json files (legacy τ² board, tau2_bench_version 0.1.3). Each submitting organisation is a separate evaluator; 'Alibaba' and 'Qwen' are the same Qwen team.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "github.com"
  },
  {
    "id": "card-meituan-longcat-flash-chat",
    "name": "Meituan LongCat-Flash-Chat model card (evaluation table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/meituan-longcat/LongCat-Flash-Chat/blob/f0b8e8d7380330f450ca1fde4ec56823a514584c/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/meituan-longcat/LongCat-Flash-Chat/blob/f0b8e8d7380330f450ca1fde4ec56823a514584c/LICENSE",
    "terms_url": "https://huggingface.co/meituan-longcat/LongCat-Flash-Chat/blob/f0b8e8d7380330f450ca1fde4ec56823a514584c/LICENSE",
    "quote": "Pinned repo front matter 'license: mit'; LICENSE at the same revision: 'MIT License Copyright (c) 2025 Meituan. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including ... to use, copy, modify, merge, publish, distribute ...'. The README is a documentation file of the repository.",
    "attribution": "Meituan LongCat Team, LongCat-Flash-Chat model card (@f0b8e8d), MIT. Selected τ²-Bench (telecom) cells measured by Meituan; percentages converted to fractions. No endorsement.",
    "scope": "Evaluation Results table, row 'τ²-Bench (telecom) (avg@4)'. Card note: 'Values marked with * are sourced from other public reports' — no telecom cell is starred, so all seven are Meituan's runs. Card note: 'DeepSeek-V3.1, Qwen3-235B-A22B, Gemini2.5-Flash, and Claude4-Sonnet are evaluated under their non-thinking mode.' Pass 8 (C): scope extended from the τ² telecom row to all unstarred cells of the Evaluation Results table on rows with existing ids, including the LongCat-Flash own column (proposed longcat-flash-chat).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card-meituan-longcat-flash-thinking",
    "name": "Meituan LongCat-Flash-Thinking model card (evaluation table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/meituan-longcat/LongCat-Flash-Thinking/blob/7c9aad1e2daceaeab3a5a03deedb8e11e7a7819f/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/meituan-longcat/LongCat-Flash-Thinking/blob/7c9aad1e2daceaeab3a5a03deedb8e11e7a7819f/LICENSE",
    "terms_url": "https://huggingface.co/meituan-longcat/LongCat-Flash-Thinking/blob/7c9aad1e2daceaeab3a5a03deedb8e11e7a7819f/LICENSE",
    "quote": "Pinned repo front matter 'license: mit' (MIT License, Copyright (c) 2025 Meituan; grant covers 'associated documentation files'). The README is a documentation file of the repository.",
    "attribution": "Meituan LongCat Team, LongCat-Flash-Thinking model card (@7c9aad1), MIT. Selected τ²-Bench-Telecom cells measured by Meituan; percentages converted to fractions. No endorsement.",
    "scope": "Evaluation Results table, row 'τ²-Bench-Telecom (Mean@4)'. Card note: 'Values marked with * are sourced from other public reports' — the GPT-5-Thinking cell (96.7*) is starred and NOT used. Pass 8 (C): scope extended to all unstarred cells on rows with existing ids, including the LongCat-Flash-Thinking own column; GPT-5-Thinking column excluded (effort unstated).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card-moonshotai-kimi-k2-instruct",
    "name": "Moonshot AI Kimi-K2-Instruct model card (instruction-model evaluation table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/README.md"
    ],
    "exact": true,
    "license": "Modified MIT License (Kimi K2)",
    "license_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/LICENSE",
    "terms_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/LICENSE",
    "quote": "LICENSE at the pinned revision: 'Modified MIT License Copyright (c) 2025 Moonshot AI Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the \"Software\"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies ... Our only modification part is that, if the Software (or any derivative works thereof) is used for any of your commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars (or equivalent in other currencies) in monthly revenue, you shall prominently display \"Kimi K2\" on the user interface of such product or service.' Card section 6: 'Both the code repository and the model weights are released under the Modified MIT License.' Conditional commercial grant (display condition above a size threshold), like the accepted LFM Open License; OWNER SIGN-OFF NEEDED (first Modified-MIT source).",
    "attribution": "Moonshot AI, Kimi-K2-Instruct model card (@fd1984e), Modified MIT License. Selected Tau2 telecom cells measured by Moonshot; percentages converted to fractions. No endorsement.",
    "scope": "Section 3 'Instruction model evaluation results', row 'Tau2 telecom' (Avg@4). Footnote: 'Data points marked with * are taken directly from the model's tech report or blog' — no Tau2 cell is starred; 'All metrics ... evaluated with an 8k output token length'; 'we employed avg@k on ... Tau2'.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "report-2601.07022v1",
    "name": "Solar Open Technical Report (arXiv:2601.07022v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2601.07022v1",
      "arxiv.org/abs/2601.07022v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2601.07022v1",
    "quote": "arXiv abstract page for 2601.07022 (v1, only version) links http://creativecommons.org/licenses/by/4.0/ as the article licence; the HTML version prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) permits reproduction and sharing, including commercially, with attribution.",
    "attribution": "Upstage Solar Team, Solar Open Technical Report, arXiv:2601.07022v1, CC BY 4.0. Table 6 'Tau 2 Telecom' cells; percentages converted to fractions. No endorsement.",
    "scope": "Table 6 (English Benchmarks), row 'Tau 2 Telecom'. Section 7.1 cites Barres et al. 2025 for 'Tau 2 (Airline/Telecom/Retail)'. Solar Open itself is not catalogued; gpt-oss-120b (high) and GLM-4.5-Air rows used; gpt-oss-120b (medium) column not used (lower setting).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2508.18669v1-mua-rl",
    "name": "MUA-RL: Multi-turn User-interacting Agent Reinforcement Learning for agentic tool use (arXiv:2508.18669v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2508.18669v1",
      "arxiv.org/abs/2508.18669v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2508.18669v1",
    "quote": "arXiv abstract page for 2508.18669v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of that version; the HTML prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) permits reproduction and sharing, including commercially, with attribution.",
    "attribution": "Weikang Zhao, Xili Wang et al. (Meituan), MUA-RL, arXiv:2508.18669v1, CC BY 4.0. τ²-bench Telecom cells measured by the authors; percentages converted to fractions. No endorsement.",
    "scope": "Table 1 'TAU2 Telecom' column, 'Existing Baselines' rows and the MUA-RL base models' 'Non-thinking' rows. Not used: the Qwen3-235B-A22B / Qwen3-30B-A3B / Qwen3-4B Non-thinking rows (24.6 / 18.4 / 17.5 are identical to Qwen's own model-card figures, so possibly copied), cold-start and MUA-RL-trained rows (not catalogue models), TCR column.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.28715v2-seataubench",
    "name": "SEATauBench: Progressively Adapting Tool-Agent-User Evaluation Into Low-Resource Southeast Asian Languages (arXiv:2606.28715v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.28715v2",
      "arxiv.org/abs/2606.28715v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.28715v2",
    "quote": "arXiv abstract page for 2606.28715v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of that version; the HTML prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) permits reproduction and sharing, including commercially, with attribution.",
    "attribution": "My Chiffon Nguyen, Aulia Adila et al., SEATauBench, arXiv:2606.28715v2, CC BY 4.0. τ²-bench Telecom cells measured by the authors; percentages converted to fractions. No endorsement.",
    "scope": "Appendix G Table 17 'English Baseline (S1) results', Telecom rows, column EN p@1 (pass@1). Telecom uses the original τ²-bench tasks ('For the telecom domain, we follow the original τ²-Bench, as its tasks were not updated in the most recent release'); airline/retail use the corrected τ³ tasks and are not used. Localized (L2) settings not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.03892v2-synthesize-and-reward",
    "name": "Synthesize and Reward: Reinforcement Learning for Multi-Step Tool Use in Live Environments (arXiv:2606.03892v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.03892v2",
      "arxiv.org/abs/2606.03892v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.03892v2",
    "quote": "arXiv abstract page for 2606.03892v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of that version; the HTML prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) permits reproduction and sharing, including commercially, with attribution.",
    "attribution": "Ibrahim Abdelaziz et al. (IBM Research), Synthesize and Reward, arXiv:2606.03892v2, CC BY 4.0. τ²-bench Telecom cells measured by the authors; percentages converted to fractions. No endorsement.",
    "scope": "Table 3 'Main results across model families', τ²-bench Telecom column, 'Base' rows only (the four instruction-tuned starting checkpoints). 'Prove'-trained rows are not catalogue models.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2602.22697v1-interactcs-rl",
    "name": "Reinforcing Real-world Service Agents: InteractCS-RL (arXiv:2602.22697v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2602.22697v1",
      "arxiv.org/abs/2602.22697v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2602.22697v1",
    "quote": "arXiv abstract page for 2602.22697v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of that version; the HTML prints 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) permits reproduction and sharing, including commercially, with attribution.",
    "attribution": "Ning Gao et al. (Meituan), InteractCS-RL, arXiv:2602.22697v1, CC BY 4.0. τ²-bench Telecom cells measured by the authors; percentages converted to fractions. No endorsement.",
    "scope": "Table 2 'Performance of our InteractCS-RL in τ²-Bench', Telecom Pass@1, base-model rows Qwen2.5-7B and Qwen2.5-14B (the paper's agents are 'based on Qwen2.5-7/14B-Instruct'). SFT/RL rows not catalogue models.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2609.13916v1-north-small-translate",
    "name": "Cohere — North Small Translate (arXiv:2609.13916v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.13916v1",
      "arxiv.org/abs/2609.13916v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.13916v1",
    "quote": "arXiv abstract page for 2609.13916v1 links http://creativecommons.org/licenses/by/4.0/; the HTML header prints 'License: CC BY 4.0'. §3: 'We evaluate all systems in an identical setup, in a clean zero-shot approach without any inference techniques', so the table cells are Cohere's own runs.",
    "attribution": "Kocmi et al. (Cohere), North Small Translate: Advanced Cost-Effective Translation, arXiv:2609.13916v1, CC BY 4.0. Table 1 cells for catalogue models; percentages divided by 100. No endorsement.",
    "scope": "Table 1 GEMBA WMT26, xCOMETxl WMT24++ and Terminology columns for catalogue models (Command A+ as Cohere's own model, by: lab; the others by: independent). Long-context (one document per language) and the internal structured-translation suite not used; DeepL, Google Translate, North Small Translate and Command A Translate are not catalogue models.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-step1x-edit-readme",
    "name": "stepfun-ai/Step1X-Edit repository README results",
    "kind": "first-party",
    "match": [
      "github.com/stepfun-ai/Step1X-Edit/blob/4cca05dfcd6c32dc810d0b4db84c2bf0682a5aeb/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/stepfun-ai/Step1X-Edit/blob/4cca05dfcd6c32dc810d0b4db84c2bf0682a5aeb/LICENSE",
    "terms_url": "https://github.com/stepfun-ai/Step1X-Edit/blob/4cca05dfcd6c32dc810d0b4db84c2bf0682a5aeb/LICENSE",
    "quote": "GitHub API reports the repository licence as Apache-2.0 (spdx_id Apache-2.0); README section '## LICENSE' at the pinned commit. Only StepFun's own-model GEdit-Bench-EN cells in the README news tables are used; the README states the judge ('G_SC, G_PQ, and G_O refer to the metrics evaluated by GPT-4.1') and the split ('GEdit-Bench-EN (Full set)').",
    "attribution": "StepFun. stepfun-ai/Step1X-Edit README @4cca05d, Apache-2.0. Selected own-model GEdit-Bench-EN G_O cells transcribed and divided by 10. No values estimated or rerun. No endorsement.",
    "scope": "Own-model rows 'Step1X-Edit (v1.0)' and 'Step1X-Edit (v1.1)' of the 'Jul 09, 2025' GEdit-Bench-EN (Full set) table, G_O column only. Not used: Qwen2.5-VL-72B (Q_*) columns, the KRIS-Bench rows (already recorded from card2-stepfun-ai-step1x-edit-v1p2), the image-only results in assets/eval_res_en.png.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "evidence_url": "https://api.github.com/repos/stepfun-ai/Step1X-Edit",
    "icon": "github.com"
  },
  {
    "id": "visual-vabench",
    "name": "VABench: A Comprehensive Benchmark for Audio-Video Generation (arXiv:2512.09299v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2512.09299v2"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2512.09299v2",
    "quote": "arXiv abs page https://arxiv.org/abs/2512.09299 (v1 10 Dec 2025, v2 6 Apr 2026) links http://creativecommons.org/licenses/by/4.0/.",
    "attribution": "D. Hua, X. Wang, B. Zeng et al. VABench: A Comprehensive Benchmark for Audio-Video Generation, arXiv:2512.09299v2, CC BY 4.0. Extracted Table 1 (T2AV) 'Visual QA' cells for end-to-end AV models and the merged video-generator cells of V+A pipelines. No values estimated or rerun. No endorsement.",
    "scope": "Benchmark authors' own runs of third-party models (Sec. 4.1 says which were accessed via APIs and which were deployed locally). Only Table 1 Visual QA used; other metrics, I2AV Table 2 and per-category Tables 5-6 not imported. Kling2.5 Turbo (720P = Standard mode) not mapped.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "part": "A"
  },
  {
    "id": "visual-kandinsky6-report",
    "name": "Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation (arXiv:2610.05608v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2610.05608v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2610.05608v1",
    "quote": "arXiv abs page https://arxiv.org/abs/2610.05608 (v1, 4 Oct 2026) links http://creativecommons.org/licenses/by/4.0/.",
    "attribution": "Kandinsky Lab (Sber AI). Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation, arXiv:2610.05608v1, CC BY 4.0. Extracted Table 11 VABench 'visual_qa' row. No values estimated or rerun. No endorsement.",
    "scope": "Authors' own VABench run of Kandinsky 6.0 Pro/Lite (lab) and of LTX 2.5 (independent: 'alongside LTX 2.5 in its two-stage Full HD configuration'). Side-by-side human results are bar charts only (not used). Memory/time tables not imported.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org",
    "part": "A"
  },
  {
    "id": "card3-open-thoughts-openthinker2-32b",
    "name": "open-thoughts/OpenThinker2-32B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/open-thoughts/OpenThinker2-32B/blob/0aefafe3ee27e073e14fb94a0563795fddf38c50/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/open-thoughts/OpenThinker2-32B/blob/0aefafe3ee27e073e14fb94a0563795fddf38c50/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0' and the card body states 'Apache 2.0 License' (section 'Intended uses & limitations'); no separate LICENSE file in the repo. Evaluation statement: 'The numbers reported in the table below are evaluated with our open-source tool Evalchemy.' (covers every row of the table, so comparator rows are the authors' own runs).",
    "attribution": "Open Thoughts team. open-thoughts/OpenThinker2-32B model card. Source: https://huggingface.co/open-thoughts/OpenThinker2-32B/blob/0aefafe3ee27e073e14fb94a0563795fddf38c50/README.md. Licence: Apache-2.0. Changes: selected numerical cells (own model and comparators evaluated by the authors with Evalchemy) transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "AIME24, AIME25, MATH500 and GPQA-D cells of the single results table, for catalogue models (OpenThinker2-32B as lab; DeepSeek-R1-Distill-Qwen-32B and S1.1-32B as independent). AMC23 and LCBv2 columns unused (no matching benchmark id).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "livebench-model-judgment-apache",
    "name": "LiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasets",
    "kind": "dataset",
    "match": [
      "huggingface.co/datasets/livebench/model_judgment/blob/5896e3b11081702c7f93f4733605fa4f5a072a11/data/leaderboard-00000-of-00001.parquet",
      "huggingface.co/datasets/livebench/math/blob/17808a1857c4504ca390b380c40ab6dc604fd8c4/data/test-00000-of-00001.parquet",
      "huggingface.co/datasets/livebench/reasoning/blob/2335f69aecfbe07126abe9b333ba04def8515c9c/data/test-00000-of-00001.parquet",
      "huggingface.co/datasets/livebench/coding/blob/204020a05f23d75e8a61d3e0373bdc9ee9f12bb0/data/test-00000-of-00001.parquet",
      "huggingface.co/datasets/livebench/language/blob/1561feebcfa3a5adaf8ecb3a0bc3b01536af5710/data/test-00000-of-00001.parquet",
      "huggingface.co/datasets/livebench/data_analysis/blob/9b73ab813303e71fa34e71598e3108df14c1222c/data/test-00000-of-00001.parquet",
      "huggingface.co/datasets/livebench/instruction_following/blob/4f7ab12f0d47848da31de92bd7cc3d7d4acfe695/data/test-00000-of-00001.parquet"
    ],
    "exact": true,
    "license": "Apache-2.0 (stated in the LiveBench datasheet)",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://github.com/LiveBench/LiveBench/blob/24364d65076429adfcba7be18af4d44fddc43dce/docs/DATASHEET.md",
    "quote": "docs/DATASHEET.md @24364d6, 'What license (if any) is it distributed under?': 'The benchmark suite is public as of June 12, 2024, distributed under the Apache License 2.0.'; 'Are there any copyrights on the data?': 'There are no copyrights on the data.'; 'Are there any fees or access/export restrictions?': 'There are no fees or restrictions.' The datasheet gives the data location as 'https://huggingface.co/livebench' and points to livebench/model_answer for raw answers. The HF cards themselves carry no licence field, and the repo LICENSE is only the copied FastChat (Apache-2.0) and LiveCodeBench (MIT) licences (GitHub reports NOASSERTION).",
    "attribution": "LiveBench (White et al., ICLR 2025), livebench/model_judgment @5896e3b with question datasets livebench/<category> @2024-10-22 revisions, Apache-2.0 per the LiveBench datasheet. Release-2024-07-26 averages recomputed from the per-question judgments with LiveBench's own aggregation; catalogue ids mapped. No endorsement.",
    "scope": "Per-question judgment scores aggregated to release 2024-07-26 overall averages for models with complete judgments. OWNER REVIEW FLAG: the licence comes from the datasheet's statement about the LiveBench 'benchmark suite'/'data' on huggingface.co/livebench, not from a card licence field; confirm this covers the judgment (score) files. The livebench.ai website tables (livebench.github.io repo, no licence) are not used. Also release 2024-08-31, only for catalogue models with no 2024-07-26 result (o1-preview, o1-mini, Qwen2.5-72B, Gemini 1.5 Pro/Flash 002, Llama-3.1-Nemotron-70B), so each model has one LiveBench release.",
    "verdict": "allowed",
    "checked": "2026-10-10"
  },
  {
    "id": "paper-2406.15877v4-bigcodebench",
    "name": "BigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2406.15877v4",
      "arxiv.org/abs/2406.15877v4",
      "arxiv.org/pdf/2406.15877v4"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2406.15877v4",
    "quote": "arXiv abs page for 2406.15877v4 (submitted Tue, 1 Apr 2025) links its licence as http://creativecommons.org/licenses/by/4.0/ (v1–v4 all CC BY 4.0; no v5 exists). Tables 6 and 7 are the authors' own runs: 'we extensively evaluate 60 state-of-the-art LLMs on BigCodeBench-Complete and 35 instruction-tuned LLMs on BigCodeBench-Instruct ... we report Pass@1 with greedy decoding for the main experiments in the zero-shot setting.'",
    "attribution": "Terry Yue Zhuo et al., 'BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions', arXiv:2406.15877v4, CC BY 4.0. Tables 6–7 greedy 'Original' pass@1 cells; converted to catalogue model ids. No endorsement.",
    "scope": "Tables 6 (Complete) and 7 (Instruct), greedy-decoding 'Original' pass@1 column, full 1,140-task set. GPT-4-Turbo row held (identity conflict with Table 4). The separate BigCodeBench leaderboard data (HF bigcode/bigcodebench-results, website repo) is NOT covered: no licence.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "opencompass-compassacademic-space-predictions",
    "name": "OpenCompass CompassAcademic leaderboard (full version) Space: committed predictions file",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/opencompass/CompassAcademic-Leaderboard-Full-Version/blob/73af091f28193d1e3e232851b940cadbb856556e/data/hf-academic-predictions.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/spaces/opencompass/CompassAcademic-Leaderboard-Full-Version/blob/73af091f28193d1e3e232851b940cadbb856556e/README.md",
    "quote": "Space README front matter at 73af091f: 'license: apache-2.0'. data/hf-academic-predictions.json (398 MB; last commit 7ff51fcd 'test', 2025-03-31) is committed in the Space repo and read by app.py (findfile_predictions). It holds OpenCompass's own IFEval (8 models) and DROP (4 models) predictions with per-model 'results'. The leaderboard's main score table is NOT covered: app.py fetches it from http://opencompass.oss-cn-shanghai.aliyuncs.com/dev-assets/hf-research/hf-academic.json (outside the repo, no licence).",
    "attribution": "OpenCompass Contributors, CompassAcademic Leaderboard (Full Version) Space, data/hf-academic-predictions.json @73af091f, Apache-2.0. IFEval prompt-level strict and DROP F1 'results' fields only; percentages converted to fractions; catalogue ids mapped. No endorsement.",
    "scope": "'results' fields of the IFEval and drop entries only. The file's cfg blocks contain third-party API credentials: never copy or republish the cfg contents.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "nvidia-ruler-readme-results",
    "name": "NVIDIA RULER repository README results table",
    "kind": "leaderboard",
    "match": [
      "github.com/NVIDIA/RULER/blob/c3f5e3b4f87f97e048793bb510a3a6b19a46bf3a/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/NVIDIA/RULER/blob/c3f5e3b4f87f97e048793bb510a3a6b19a46bf3a/LICENSE",
    "terms_url": "https://github.com/NVIDIA/RULER/blob/c3f5e3b4f87f97e048793bb510a3a6b19a46bf3a/LICENSE",
    "quote": "Root LICENSE at the pinned commit is the Apache License 2.0 (GitHub API: Apache-2.0); the README with the results table is a committed file of that repository. README: 'We benchmark 17 open-source models ... Here are our main results.' Notes list the rows reported by model authors (Jamba-1.5-large, FILM-7B, Qwen2.5-1M, Qwen3, EXAONE 4.0); GradientAI rows were added in commit e8a25c3 'add results from GradientAI'.",
    "attribution": "NVIDIA, RULER (Hsieh et al., 2024), github.com/NVIDIA/RULER README @c3f5e3b, Apache-2.0. 64K and 128K columns of NVIDIA's own rows; percentages converted to fractions; catalogue ids mapped. No endorsement.",
    "scope": "README main table, 64K and 128K columns, rows without an asterisk and not from the GradientAI commit, only where the evaluated length is within the row's claimed context length.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "paper-2410.07112v2-vhelm",
    "name": "VHELM: A Holistic Evaluation of Vision Language Models (Stanford CRFM), arXiv 2410.07112v2",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2410.07112v2",
      "arxiv.org/abs/2410.07112v2",
      "arxiv.org/pdf/2410.07112v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2410.07112v2",
    "quote": "arXiv abs page for 2410.07112v2 (submitted 24 Oct 2024) links http://creativecommons.org/licenses/by/4.0/. §4: 'We evaluate 22 recent models from 6 developers, as listed in Table A2 ... In addition to implementing consistent adaptation and evaluation methods across all the models'; §3: 'In this study, we use only zero-shot prompting'.",
    "attribution": "Tony Lee et al., 'VHELM: A Holistic Evaluation of Vision Language Models', arXiv:2410.07112v2, CC BY 4.0. Table 4 MMMU (EM) and MathVista (EM) cells; catalogue ids mapped. No endorsement.",
    "scope": "Table 4, MMMU (EM) and MathVista (EM) columns for catalogue API models. Judge-scored (Prometheus-Vision) columns and mean win rates (pool-dependent) not used. The live HELM/VHELM leaderboard data (GCS bucket, no licence) is NOT covered.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "speech-stt-danish-asr-ryeai-mit",
    "name": "Danish Speech-to-Text Leaderboard (RyeAI) · results dataset",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/datasets/RyeAI/danish-asr-leaderboard/blob/344e5a7047d9b5f3b1c8248050ae5e1dba186f43/data/results.parquet"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://github.com/Rye-A1/danish-asr-leaderboard/blob/f4a0ddc2d6703c43130788507a89231a44f32d1d/LICENSE",
    "terms_url": "https://huggingface.co/datasets/RyeAI/danish-asr-leaderboard/blob/344e5a7047d9b5f3b1c8248050ae5e1dba186f43/README.md",
    "quote": "Results dataset RyeAI/danish-asr-leaderboard @344e5a7: card 'license: mit'. Space RyeAI/danish-asr-leaderboard @feb3f84: 'license: mit' with the same numbers baked into the committed leaderboard.json. Harness github.com/Rye-A1/danish-asr-leaderboard @f4a0ddc README: 'every model is transcribed and scored the same way on the same five public Danish test sets' ... 'License: MIT'.",
    "attribution": "RyeAI, Danish Speech-to-Text Leaderboard (huggingface.co/datasets/RyeAI/danish-asr-leaderboard @344e5a7, MIT). mean_wer column; percentages converted to fractions; catalogue ids mapped. No endorsement.",
    "scope": "mean_wer of catalogue models. OWNER REVIEW FLAG: the README invites outside contributors to 'Run the eval and open a PR with the result JSON', and rows carry no submitter field; the imported rows are general-purpose models (Whisper, NVIDIA, Mistral, Qwen, Meta, OpenAI/ElevenLabs APIs) run with the published harness, and per-sample hypotheses are published under outputs/ for re-scoring.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "speech-stt-dutch-asr-tvosch-apache",
    "name": "Dutch ASR Leaderboard (tvosch) · committed result files",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/results/CohereLabs__cohere-transcribe-03-2026.json",
      "huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/results/ElevenLabs-scribe_v2.json",
      "huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/results/Qwen__Qwen3-ASR-1.7B.json",
      "huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/results/mistralai__Voxtral-Mini-4B-Realtime-2602.json",
      "huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/results/nvidia__parakeet-tdt-0.6b-v3.json",
      "huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/results/openai__whisper-large-v3.json",
      "huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/results/voxtral-mini-2602.json"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/spaces/tvosch/Dutch-ASR-Leaderboard/blob/939431c9206d1e7676347a95d13317f9a3226181/README.md",
    "quote": "Space card at 939431c: 'license: apache-2.0'; results/<model>.json files are committed in the Space and read by app.py; README: 'Local models are benchmarked on H100 HBM2e GPUs' and 'How to Submit: 1. Run eval.py to generate results JSON 2. Add your results/<model>.json file 3. Open a PR'.",
    "attribution": "tvosch, Dutch ASR Leaderboard Space (@939431c, Apache-2.0). Three-set average WER recomputed as app.py does; catalogue ids mapped. No endorsement.",
    "scope": "Seven result files of catalogue models. OWNER REVIEW FLAG: community-PR leaderboard; files carry per-sample outputs but no submitter field. Vendor-looking entries (Murmel, Resonate-1) and the community fine-tune Qwen3-ASR-1.7B-NL are not imported.",
    "verdict": "allowed",
    "checked": "2026-10-09"
  },
  {
    "id": "card3-kwaipilot-kat-coder-v2-5-dev",
    "name": "Kwaipilot/KAT-Coder-V2.5-Dev model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev/blob/7be56fe773e72b6f5ca93c1ae45d828ddb893922/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev/blob/7be56fe773e72b6f5ca93c1ae45d828ddb893922/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'. Note 1 under the benchmark table: 'All metrics presented in the table are reproduced in-house: we download the public model checkpoints, deploy them via vLLM or SGLang, and evaluate under a unified standardized pipeline. No officially reported results of the respective models are directly adopted in this table.' So every column is Kwaipilot's own measurement.",
    "attribution": "Kwaipilot (Kuaishou). Kwaipilot/KAT-Coder-V2.5-Dev model card. Source: https://huggingface.co/Kwaipilot/KAT-Coder-V2.5-Dev/blob/7be56fe773e72b6f5ca93c1ae45d828ddb893922/README.md. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100; Terminal-Bench 2.1 per-harness (Terminus-2) values used. No endorsement.",
    "scope": "Benchmark performance table: KAT-Coder-V2.5-Dev (by: lab) and catalogued comparator columns Qwen3.5-27B, Qwen3.6-35B-A3B, Gemma 4 31B, Qwen3.5-35B-A3B, Gemma 4 26B-A4B, Qwen3-Coder-30B-A3B (by: independent, authors' in-house runs). Ornith-1.0-35B not catalogued. PinchBench and KAT-Code-Bench (internal) have no id. Pass 8 (C): the 'Ornith-1.0-35B' column (in-house run, card note 1) is now also used for the proposed ornith-1-0-35b (by: independent).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-deepseek-ai-deepseek-v4-flash-vision-exp",
    "name": "deepseek-ai/DeepSeek-V4-Flash-Vision-Exp model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/README.md",
    "terms_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'; '## License: This repository is licensed under the MIT License.' Own-lab columns (DeepSeek-V4-Flash-Vision-Exp, DeepSeek-V4-Flash-0731) only; the Opus-4.8 column is not used.",
    "attribution": "DeepSeek-AI. deepseek-ai/DeepSeek-V4-Flash-Vision-Exp model card. Source: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Own-lab columns of the introduction table on benchmarks with existing ids.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-xiaomimimo-mimo-v2-6-flash-rl",
    "name": "XiaomiMiMo/MiMo-V2.6-Flash-RL model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/5711b268169967567844e1e560e8a3966da959b1/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/5711b268169967567844e1e560e8a3966da959b1/README.md",
    "terms_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/5711b268169967567844e1e560e8a3966da959b1/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit' (repo licence tag mit). Own-lab columns of the '## 3. Evaluation Results' table (MiMo-V2.6 Pro, MiMo-V2.6 Flash, MiMo-V2.5 Pro) only. The Claude Opus 5 / GPT-5.6 Sol / Claude Fable 5 columns are NOT used: no statement that Xiaomi measured them, and several cells equal the vendors' own published figures (e.g. GPT-5.6 Sol Terminal Bench 2.1 88.8 = OpenAI's 0.888; Claude Fable 5 Terminal Bench 2.1 84.3 = Anthropic's 0.843; Claude Opus 5 Toolathlon-Verified 80.6 = Anthropic's 0.806). GDPval-AA (Artificial Analysis) and CyberGym (environments modified by Xiaomi, report §5.2 footnote 1) rows not used.",
    "attribution": "Xiaomi MiMo Team. XiaomiMiMo/MiMo-V2.6-Flash-RL model card. Source: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/5711b268169967567844e1e560e8a3966da959b1/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Own-lab columns of the evaluation table; benchmark rows mapped to existing ids (deepswe-v1-1, toolathlon-verified, agents-last-exam-unspecified, terminal-bench-4, terminal-bench-2.1, osworld-verified, automationbench-v1-0-6, exploitgym, sec-bench-pro).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-xiaomimimo-mimo-v2-6-distill-qwen-9b",
    "name": "XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B/blob/2367e865d009c13ac81713a2878291d33ab28177/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B/blob/2367e865d009c13ac81713a2878291d33ab28177/README.md",
    "terms_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B/blob/2367e865d009c13ac81713a2878291d33ab28177/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit'. Own-model column 'MiMo-V2.6-Distill-Qwen-9B (SFT)' only; the Qwen3.5-9B column is held for the owner (see held_for_owner).",
    "attribution": "Xiaomi MiMo Team. XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B model card. Source: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B/blob/2367e865d009c13ac81713a2878291d33ab28177/README.md. Licence: MIT. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Own-model SFT column of the evaluation table (public benchmarks with existing ids).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-ornith-ai-ornith-1-5-35b-a3b",
    "name": "ornith-ai/Ornith-1.5-35B-A3B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/10fbf86fed7ecee4a061f8b499a618f46001cac1/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://opensource.org/licenses/MIT",
    "terms_url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/10fbf86fed7ecee4a061f8b499a618f46001cac1/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: mit' (the Ornith 1.0 card: 'Licence: MIT licensed, globally accessible, and free from regional limitations'). Own-lab Ornith 1.5 and Ornith 1.0 columns only; comparator columns (Qwen3.6-35B-A3B, Gemma-4-31B, Muse-Glimmer-30B, Qwen3.5-397B) are not used: the card does not say the authors measured them, and several cells equal those makers' published numbers.",
    "attribution": "DeepReinforce (ornith-ai). Ornith-1.5-35B-A3B model card. Source: https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/10fbf86fed7ecee4a061f8b499a618f46001cac1/README.md. Licence: MIT. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Benchmark table: Ornith-1.5-35B-A3B and Ornith-1.0-35B-A3B columns (by: lab) on 12 rows with existing ids. Claude Code harness Terminal-Bench row, DeepSWE (version not stated), Frontier-Bench, SWE Atlas and ClawEval (version/metric not stated) not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "first-party-poolside-scores",
    "name": "Poolside vendor-reported benchmark scores for its own Laguna models (Hugging Face model cards)",
    "match": [
      "huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/README.md#benchmark-results",
      "huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/README.md#benchmark-results"
    ],
    "kind": "first-party",
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/README.md",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "exact": true,
    "icon": "huggingface.co",
    "attribution": "Poolside, vendor-reported results for Poolside's own Laguna models from its pinned Hugging Face cards; individual figures hand-recorded 2026-10-10; percentages converted to fractions. Lab-reported. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. Cards are tagged openmdw-1.1; recorded under rule (b) rather than as licensed copies (same treatment as first-party-arcee-scores). Match URLs carry the '#benchmark-results' fragment so the score rows stay separate from the first-party-poolside price/doc entry. Comparison rows on these cards are other makers' published or third-party (incl. Artificial Analysis) figures and are not used."
  },
  {
    "id": "card3-openbmb-minicpm5-2b",
    "name": "openbmb/MiniCPM5-2B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/openbmb/MiniCPM5-2B/blob/f97400052a43d642bbc6e9975e2397e3ae6a6b52/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/openbmb/MiniCPM5-2B/blob/f97400052a43d642bbc6e9975e2397e3ae6a6b52/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0'. Table footnote 2: 'Scores marked † come from the official Artificial Analysis release; all others are reproduced internally.' Only unmarked cells are used (the † cells are Artificial Analysis figures and are excluded).",
    "attribution": "OpenBMB. openbmb/MiniCPM5-2B model card. Source: https://huggingface.co/openbmb/MiniCPM5-2B/blob/f97400052a43d642bbc6e9975e2397e3ae6a6b52/README.md. Licence: Apache-2.0. Changes: selected unmarked numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "'Evaluation Results of MiniCPM5-2B and Baselines' table: MiniCPM5-2B (by: lab) and OpenBMB's internally reproduced columns LFM2.5-2.6B, Qwen3.5-2B, Gemma-4-E2B-it, Qwen3.5-4B, granite-4.2-3B, Nemotron-3-Nano-4B, Gemma-4-E4B-it, LFM2.5-8B-A1B (by: independent) on 21 rows with existing ids. All † cells (SciCode, HLE, GPQA-Diamond, AA-LCR, Terminal-Bench v2.1 and some τ²/τ³/GDPval cells) excluded as Artificial Analysis figures; NoLiMa, LongBenchPro, BrowseComp Top100, GAIA Text-103, GDPval-AA, Claw-Gym, WildClaw, QwenClaw have no id.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-lgai-exaone-k-exaone-2-0-750b-a37b",
    "name": "LGAI-EXAONE/K-EXAONE-2.0-750B-A37B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/LGAI-EXAONE/K-EXAONE-2.0-750B-A37B/blob/fec0c8dd25fdd20ffd09869d5af5277966d5523d/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/LGAI-EXAONE/K-EXAONE-2.0-750B-A37B/blob/fec0c8dd25fdd20ffd09869d5af5277966d5523d/LICENSE",
    "terms_url": "https://huggingface.co/LGAI-EXAONE/K-EXAONE-2.0-750B-A37B/blob/fec0c8dd25fdd20ffd09869d5af5277966d5523d/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0' with 'license_link: LICENSE'; the card's License section: 'The model is licensed under Apache License 2.0' and 'We also release K-EXAONE 2.0 under the Apache license 2.0 so that the broader AI ecosystem can inspect, deploy, and build upon it.'",
    "attribution": "LG AI Research. LGAI-EXAONE/K-EXAONE-2.0-750B-A37B model card. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Evaluation Results table: K-EXAONE 2.0 and K-EXAONE (236B) columns (LG's own models, by lab) on MMLU-Pro, GPQA-Diamond, HLE (text-only), AIME 2026, HMMT Feb 2026, IMO-AnswerBench, SWE-bench Verified, Terminal-Bench 2.1, τ3-Banking, IFEval, IFBench; plus the comparator cells that the technical report (arXiv 2608.04505v1, Table 6) leaves without an asterisk, i.e. LG's own internal runs ('when official scores are unavailable, we evaluate them in our internal environment'): GLM-5.1 MMLU-Pro/SWE-bench Verified/IFEval, DeepSeek V4 Pro (max, April preview) AIME 2026/IFEval, Qwen3.5-397B HMMT Feb 2026. Asterisked (copied official) cells excluded. Not used: SciCode and Claw-Eval (card and report values differ for K-EXAONE 2.0), AA-LCR, OpenAI-MRCR, Ko-LongBench, Korean rows, MMMLU/GlobalMMLU-Lite/PolyMath (nine-language subsets), safety rows.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co",
    "evidence_url": "https://arxiv.org/html/2608.04505v1"
  },
  {
    "id": "speech-stt-cose-e-ccby",
    "name": "CoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings (arXiv:2609.35645v1), Table 2",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.35645v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.35645v1",
    "quote": "arXiv abs page for 2609.35645v1 links http://creativecommons.org/licenses/by/4.0/ (arXiv OAI metadata license: http://creativecommons.org/licenses/by/4.0/). §4: 'We evaluate eight frontier models covering both proprietary and open-weight across five language pairs ... all models use native auto-language detection without explicit language parameter'; Appendix C.3 Table 8 'ASR systems and decoding parameters. All models were run with no language-ID passed.' Authors are from ServiceNow AI Research and Qualcomm Technologies (no evaluated ASR system is theirs).",
    "attribution": "Shama Gupta, Hoang H. Nguyen, Chelsea Huang, Lindsay Devon Brin, Fanny Riols et al., 'CoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings', arXiv:2609.35645v1 (2026), CC BY 4.0. Changes: Table 2 European-pair WER cells averaged per model. No endorsement.",
    "scope": "Table 2 WER cells for the four European pairs of six catalogue STT systems (Whisper row stored invalid). SWER/AER, ZH/EN and the Gemini/Qwen3-Omni rows not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "evidence_url": "https://arxiv.org/abs/2609.35645v1",
    "icon": "arxiv.org"
  },
  {
    "id": "speech-stt-iberian-bench-ccby",
    "name": "Benchmarking Automatic Speech Recognition Tools for Iberian Languages (arXiv:2609.36920v1), Table 3",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.36920v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.36920v1",
    "quote": "arXiv OAI metadata and abs page for 2609.36920v1: license http://creativecommons.org/licenses/by/4.0/. Abstract: 'We benchmark eleven systems, ten open-weight models and one commercial API, across five Iberian languages ... with German and Turkish as controls.' §2.4: 'All models run on a single NVIDIA GeForce RTX 3090 (24 GB VRAM) ... For models with configurable delay (e.g., Voxtral-Mini-4B-Realtime), we use the default setting.' Authors from Telefónica Innovación Digital and Universidad Autónoma de Madrid; none of the evaluated systems is theirs.",
    "attribution": "Fernando López, Pablo Gómez, David Solans, Paulo Villegas, Jordi Luque, 'Benchmarking Automatic Speech Recognition Tools for Iberian Languages', arXiv:2609.36920v1 (2026), CC BY 4.0. Table 3 macro WER, % divided by 100. No endorsement.",
    "scope": "Table 3 macro-WER column for all eleven systems (all catalogued). RTF, error-type rates and per-language/per-sex figures not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "evidence_url": "https://arxiv.org/abs/2609.36920v1",
    "icon": "arxiv.org"
  },
  {
    "id": "speech-stt-mu-bench-ccby",
    "name": "mu-bench: A Multilingual Utterance Transcription Benchmark (arXiv:2609.32082v1), Table 5",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.32082v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.32082v1",
    "quote": "arXiv OAI metadata for 2609.32082v1: license http://creativecommons.org/licenses/by/4.0/. §4.1: 'We tested six commercial APIs through their batch endpoints and three through their streaming endpoints (Table 5) ... the locale is passed as a language hint, and all other parameters are left at their defaults.' Authors from UC Berkeley and Sierra AI (a voice-agent company, not an ASR vendor). The public leaderboard at research.sierra.ai/mubench is not used; only the paper's Table 5.",
    "attribution": "Andrea Li, Soham Ray et al., 'mu-bench: A Multilingual Utterance Transcription Benchmark', arXiv:2609.32082v1 (2026), CC BY 4.0. Table 5 batch 'all' UER, % divided by 100. No endorsement.",
    "scope": "Table 5 batch-mode 'all' UER for Google Chirp-3, ElevenLabs Scribe v2, OpenAI GPT-4o Mini Transcribe and Deepgram Nova-3. Streaming rows, Azure Speech and xAI Grok STT rows not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "evidence_url": "https://arxiv.org/abs/2609.32082v1",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-inworld",
    "name": "Inworld official pricing page",
    "match": [
      "inworld.ai/pricing"
    ],
    "kind": "first-party",
    "license": "Factual list price (not licensed content)",
    "terms_url": "https://inworld.ai/pricing",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "exact": true,
    "icon": "inworld.ai",
    "attribution": "Inworld official pricing page, observed 2026-10-09",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "card3-meituan-longcat-longcat-audiodit-3-5b",
    "name": "meituan-longcat/LongCat-AudioDiT-3.5B model card (Seed benchmark table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/meituan-longcat/LongCat-AudioDiT-3.5B/blob/5d3e6dc007872d756c3a69b44951598d7d9bdf6d/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/meituan-longcat/LongCat-AudioDiT-3.5B/blob/5d3e6dc007872d756c3a69b44951598d7d9bdf6d/LICENSE",
    "terms_url": "https://huggingface.co/meituan-longcat/LongCat-AudioDiT-3.5B/blob/5d3e6dc007872d756c3a69b44951598d7d9bdf6d/README.md",
    "quote": "README front matter at the pinned revision: 'license: mit'; section 'License Agreement': 'This repository, including both the model weights and the source code, is released under the MIT License.' Repository LICENSE file (1,064 bytes) at the same revision. Only the two LongCat-AudioDiT rows (1B, 3.5B) of the 'Experimental Results on Seed Benchmark' table are used; the evaluation setup is from the same authors' arXiv:2603.29339v1 §5.1 (CC BY-SA 4.0, used for setup/identity only). Comparator rows not used: the paper says 'The results of other methods are taken from the original paper or, if open-sourced, evaluated by us', without marking which.",
    "attribution": "Meituan LongCat Team. meituan-longcat/LongCat-AudioDiT-3.5B model card, https://huggingface.co/meituan-longcat/LongCat-AudioDiT-3.5B/blob/5d3e6dc007872d756c3a69b44951598d7d9bdf6d/README.md, MIT. Changes: own-model cells transcribed, percentages divided by 100. No endorsement.",
    "scope": "Seed-TTS test-zh CER / test-en WER / WavLM SIM for LongCat-AudioDiT-1B and -3.5B (same-lab rows).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "speech-vibevoice-asr-streaming-report-ccby",
    "name": "VibeVoice-ASR-Streaming Technical Report (arXiv:2609.02812v2), Table 3 single-speaker recognition",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.02812v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.02812v2",
    "quote": "arXiv abstract page for 2609.02812v2 links http://creativecommons.org/licenses/by/4.0/. Table 3 note: 'All rows are our own measurements under a single normalization. X-ASR is evaluated in its chunk-1920ms streaming configuration, Voxtral-Mini-4B-Realtime at transcription_delay_ms=2400 (2.4 s), its longest configurable delay, and Nemotron-3.5-ASR at att_context_size=[56,13] (1.12 s), in each case the released setting closest to our chunk sizes'.",
    "attribution": "Microsoft Research, 'VibeVoice-ASR-Streaming Technical Report', arXiv:2609.02812v2 (2026), CC BY 4.0. Table 3 WER cells, % / 100. No endorsement.",
    "scope": "Table 3 LibriSpeech test-clean/test-other and GigaSpeech WER for the authors' released 22-frame 7B model and their own runs of Nemotron-3.5-ASR 0.6B and Voxtral-Mini-4B-Realtime. Meeting/MLC tables (no matching benchmark ids) and API rows not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "arxiv.org"
  },
  {
    "id": "card3-microsoft-vibevoice-asr-bitnet",
    "name": "microsoft/VibeVoice-ASR-BitNet model card (Accuracy table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/microsoft/VibeVoice-ASR-BitNet/blob/66e78021ab8f5f06133d1ab421ba4d348bda97c9/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://opensource.org/licenses/MIT",
    "terms_url": "https://huggingface.co/microsoft/VibeVoice-ASR-BitNet/blob/66e78021ab8f5f06133d1ab421ba4d348bda97c9/README.md",
    "quote": "README front matter at the pinned revision: 'license: mit'; '## License: This project is licensed under the MIT License.' The same table is printed in arXiv:2607.21075v2 Table 4 (CC BY 4.0), whose caption identifies the comparators (Parakeet-TDT-0.6B-v2, Whisper Large-v3, SenseVoice-small, FunASR-Nano) without saying who ran them; those columns are not used.",
    "attribution": "Microsoft Research. microsoft/VibeVoice-ASR-BitNet model card, https://huggingface.co/microsoft/VibeVoice-ASR-BitNet/blob/66e78021ab8f5f06133d1ab421ba4d348bda97c9/README.md, MIT. Changes: same-lab WER cells transcribed, % / 100. No endorsement.",
    "scope": "VibeVoice-ASR-7B and VibeVoice-ASR-BitNet columns (by: lab) and, since pass 12, the Parakeet / Whisper / SenseVoice / FunASR comparator columns (by: independent; identified as Parakeet-TDT-0.6B-v2, Whisper Large-v3, SenseVoice-small, FunASR-Nano by arXiv:2607.21075v2 Table 4, CC BY 4.0): FLEURS en/zh, LibriSpeech clean/other, VoxPopuli.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-nineninesix-gepard-1-0",
    "name": "nineninesix/gepard-1.0 model card (Seed-TTS-eval benchmark table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/nineninesix/gepard-1.0/blob/440c3a27797e98673c8e56f0569a86c09ea5de34/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://opensource.org/licenses/Apache-2.0",
    "terms_url": "https://huggingface.co/nineninesix/gepard-1.0/blob/440c3a27797e98673c8e56f0569a86c09ea5de34/README.md",
    "quote": "README front matter at the pinned revision: 'license: apache-2.0'; Model Details: 'License: Apache 2.0 (the codec has NVIDIA Open Model License Agreement)'. Benchmarks: 'Measured on the public Seed-TTS-eval — 1088 paired prompts, identical UUIDs and texts across every model.' The accompanying report arXiv:2609.04222v1 (CC BY 4.0) §6.3 prints the same table and states the six baselines were 'our own runs on identical inputs, not a compilation of third-party numbers'.",
    "attribution": "nineninesix.ai (Denis Pavlov, Ulanbek Abdurazakov, Nursultan Bakashov). nineninesix/gepard-1.0 model card, Apache-2.0; GEPARD technical report arXiv:2609.04222v1, CC BY 4.0. Changes: WER, SIM, UTMOS cells transcribed; UTMOS stored (x-1)/4. No endorsement.",
    "scope": "Seed-TTS test-en WER (Whisper, size unstated), WavLM SIM and UTMOS for Gepard 1.0 and the authors' own runs of VoxCPM2, Fish-S2 (s2-pro) and OmniVoice. Qwen3-TTS, Chatterbox (size/variant not identified) and Echo-TTS (not catalogued) not used; NISQA columns have no benchmark id; Appendix C non-cloning runs not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-moondream-parakeet-cards",
    "name": "Moondream parakeet-ultra and parakeet-redux model cards (Benchmarks sections)",
    "kind": "first-party",
    "match": [
      "huggingface.co/moondream/parakeet-ultra/blob/73175eb7aeb0d82f1e2a6b53b3aabc10a90bcd0b/README.md",
      "huggingface.co/moondream/parakeet-redux/blob/2bf128600aac4b16946f7ed8372e56117fe5e23b/README.md"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://huggingface.co/moondream/parakeet-ultra/blob/73175eb7aeb0d82f1e2a6b53b3aabc10a90bcd0b/README.md",
    "quote": "Both cards' front matter at the pinned revisions: 'license: cc-by-4.0'. Benchmarks: 'Word error rate in percent, lower is better ... Both models are scored on the same files with the Open ASR Leaderboard's own pipeline as of September 2026 (its normalizers and compound-merging alignment, with the FLEURS references prepared as the leaderboard's text column is). parakeet-ultra [parakeet-redux] runs in Photon on an NVIDIA GPU, the original in NeMo in bf16.' The parakeet-tdt-0.6b-v3 column is Moondream's own run and is identical in both cards.",
    "attribution": "Moondream (moondream.ai). moondream/parakeet-ultra and moondream/parakeet-redux model cards, CC BY 4.0. Changes: 'average' WER cells of the Open ASR (7 sets), FLEURS (25 languages) and TED-LIUM long-form tables, percent divided by 100. No endorsement.",
    "scope": "Pooled as one maker's evaluation (same stated pipeline, same parakeet-tdt-0.6b-v3 reference column). Own models by: lab; NVIDIA parakeet-tdt-0.6b-v3 column by: independent, taken once (from the parakeet-ultra card). Per-set and per-language cells, the AA-WER-style business-speech set and the MUSAN noise conditions are not used.",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-ai-sage-gigaam-multilingual",
    "name": "ai-sage/GigaAM-Multilingual model card (Model Performance table)",
    "kind": "first-party",
    "match": [
      "huggingface.co/ai-sage/GigaAM-Multilingual/blob/2f8a57144e6ec3adfd32fe0484d9ea9913305bc8/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://opensource.org/licenses/MIT",
    "terms_url": "https://huggingface.co/ai-sage/GigaAM-Multilingual/blob/2f8a57144e6ec3adfd32fe0484d9ea9913305bc8/README.md",
    "quote": "README front matter at the pinned revision: 'license: mit'. 'Model Performance: Word Error Rate (%) on Common Voice (CV), FLEURS, and internal in-the-wild test sets. Utterances longer than 30 s and references containing digits are excluded; references/hypotheses are normalized (lowercasing, punctuation removal, numerals→words); greedy decoding.'",
    "attribution": "SaluteDevices / Sber GigaAM team. ai-sage/GigaAM-Multilingual model card, MIT. Changes: English CV and FLEURS cells of the two released CTC variants, percent divided by 100. No endorsement.",
    "scope": "Own-model columns GigaAM Multilingual (ctc) and GigaAM Multilingual Large (large_ctc), English rows only (no benchmark ids exist for the Russian/Kazakh/Kyrgyz/Uzbek or internal sets). Comparator columns (Omnilingual 1B LLM, SeamlessM4T large v2, Whisper large v3) not used: the card does not say the authors ran them (held for owner).",
    "verdict": "allowed",
    "checked": "2026-10-09",
    "icon": "huggingface.co"
  },
  {
    "id": "first-party-lm-provers",
    "name": "LM Provers official model cards and configuration",
    "match": [
      "huggingface.co/lm-provers/QED-Nano/blob/1016ddef8dd40552b97216ce34e3fece2fffa2bb/config.json"
    ],
    "kind": "first-party",
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/lm-provers/QED-Nano/blob/1016ddef8dd40552b97216ce34e3fece2fffa2bb/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "LM Provers official model cards and configuration, observed 2026-10-10",
    "verdict": "allowed",
    "checked": "2026-10-10"
  },
  {
    "id": "first-party-openbmb",
    "name": "OpenBMB official model cards and configuration",
    "match": [
      "huggingface.co/openbmb/MiniCPM3-4B/blob/d6b14ddaefdb11c624dd75c3c779549bc90b08cb/README.md"
    ],
    "kind": "first-party",
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/openbmb/MiniCPM3-4B/blob/d6b14ddaefdb11c624dd75c3c779549bc90b08cb/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "OpenBMB official model cards and configuration, observed 2026-10-10",
    "verdict": "allowed",
    "checked": "2026-10-10"
  },
  {
    "id": "first-party-simplescaling",
    "name": "Simple Scaling (s1) official model cards and configuration",
    "match": [
      "huggingface.co/simplescaling/s1.1-32B/blob/7b641db5bccde475f38ab3ee8345b4521ab12d64/config.json"
    ],
    "kind": "first-party",
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/simplescaling/s1.1-32B/blob/7b641db5bccde475f38ab3ee8345b4521ab12d64/config.json",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "Simple Scaling (s1) official model cards and configuration, observed 2026-10-10",
    "verdict": "allowed",
    "checked": "2026-10-10"
  },
  {
    "id": "first-party-deepreinforce",
    "name": "DeepReinforce (Ornith) official model cards and configuration",
    "match": [
      "huggingface.co/ornith-ai/Ornith-1.0-35B/blob/5df2ed3f675c7beaa490328cc70bb573b65fb660/README.md"
    ],
    "kind": "first-party",
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B/blob/5df2ed3f675c7beaa490328cc70bb573b65fb660/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "DeepReinforce (Ornith) official model cards and configuration, observed 2026-10-10",
    "verdict": "allowed",
    "checked": "2026-10-10"
  },
  {
    "id": "first-party-naver",
    "name": "NAVER Cloud official model cards and configuration",
    "match": [
      "huggingface.co/naver-hyperclovax/HyperCLOVAX-SEED-Think-32B/blob/3c2b6338613c4e5ede30788d868363125a262934/README.md"
    ],
    "kind": "first-party",
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/naver-hyperclovax/HyperCLOVAX-SEED-Think-32B/blob/3c2b6338613c4e5ede30788d868363125a262934/README.md",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "NAVER Cloud official model cards and configuration, observed 2026-10-10",
    "verdict": "allowed",
    "checked": "2026-10-10"
  },
  {
    "id": "first-party-ai21",
    "name": "AI21 Labs official docs",
    "match": [
      "docs.ai21.com/docs/jamba-foundation-models"
    ],
    "kind": "first-party",
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://docs.ai21.com/docs/jamba-foundation-models",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. Extended 2026-10-06 (llm-context lane) to the maker's own pinned model card / official technical report / official docs stating the context window of its own model; never used for benchmark scores.",
    "attribution": "AI21 Labs official docs, observed 2026-10-10",
    "verdict": "allowed",
    "checked": "2026-10-10"
  },
  {
    "id": "card3-canberkkkkkk-ema-lightning",
    "name": "EMA Lightning model card (canberkkkkkk/ema-lightning), Benchmarks: Freya-TR-Eval",
    "kind": "first-party",
    "match": [
      "huggingface.co/canberkkkkkk/ema-lightning/blob/7a6ba1ad216bb2f1da9863f80ac8770a6a807632/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/canberkkkkkk/ema-lightning/blob/7a6ba1ad216bb2f1da9863f80ac8770a6a807632/README.md",
    "quote": "README front matter at the pinned revision: 'license: apache-2.0'; card text: 'A 5.6M-parameter DiT acoustic model and a 3M-parameter vocoder ... Apache 2.0.' Benchmarks section: 'Our runs (every row without †): all 495 sentences, 3 seeds or takes, raw text in, an 8 kHz band, Whisper large-v3 with beam 5, and Freya's text normalisation. ElevenLabs uses its default voice (George) with default settings. Gemini uses the voice Kore.' and '† marks numbers as reported on the Anka TTS model card.'",
    "attribution": "EMA Lightning model card by canberkkkkkk (Hugging Face), Apache-2.0. Changes: WER % divided by 100; only rows without † used.",
    "scope": "Benchmarks table (Freya-TR-Eval, all 495 sentences): WER of the rows without † (EMA Lightning, Trendyol-TTS, Gemini 3.8 Flash-Lite, Gemini 3.8 Flash, ElevenLabs v4, ElevenLabs v4 Turbo, Antalia 1). † rows (copied from the Anka TTS card) not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-cloud0day3-antalia-mini",
    "name": "Antalia-2 Mini model card (cloud0day3/antalia-mini), public Freya-TR-Eval re-run table",
    "kind": "first-party",
    "match": [
      "huggingface.co/cloud0day3/antalia-mini/blob/1e8166a7436f3e11f7f03b339258ec30f366db60/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/cloud0day3/antalia-mini/blob/1e8166a7436f3e11f7f03b339258ec30f366db60/README.md",
    "quote": "README front matter at the pinned revision: 'license: apache-2.0'. 'We also keep an independent re-run of Freya-TR-Eval for every Turkish system we could run, at speech.patientdesk.ai/benchmarks (scripts and per-clip transcripts: PatientdeskAI/freya-tr-eval-rerun). It uses EMA Lightning's published eval.py for scoring, 3 seeded takes per system, and one NVIDIA L4 for speed'. The scripts dataset PatientdeskAI/freya-tr-eval-rerun (license: apache-2.0, pinned d5edf41) names each system's exact checkpoint (harness/systems/run_*.py, synth_api.py).",
    "attribution": "Antalia-2 Mini model card by Sezgin Saygili, Emre Kaplaner, Oncel Ozgul and Fikri San Koktas (Patientdesk.ai), Apache-2.0. Changes: WER % divided by 100.",
    "scope": "'On the public benchmark page' table (15 systems): WER column for systems with a pinned catalogue identity. Not used: Anka TTS, Alania-1/2 (no public checkpoint pinned), Chatterbox Multilingual (pip package version not pinned), and the card's first table (Antalia-2 Mini vs EMA vs VoxCPM2 teacher, a separate harness).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "visual-multitextedit-ccby",
    "name": "MultiTextEdit: Benchmarking Cross-Lingual Degradation in Text-in-Image Editing (arXiv:2605.08163v2), Table 3",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.08163v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.08163v2",
    "quote": "arXiv abs page for 2605.08163 (v2, 18 May 2026) links http://creativecommons.org/licenses/by/4.0/; HTML header 'License: CC BY 4.0'. Sec. 5: 'We evaluate a diverse set of open-source and proprietary image editing systems on MultiTextEdit. All open-source model inference and the automatic evaluation pipeline are conducted on eight NVIDIA A800 GPUs. For each evaluated system, we follow the official inference pipeline and adopt the highest-quality inference settings recommended by the provider ... For proprietary systems, we access their official APIs'. Authors (Harbin Institute of Technology) built none of the evaluated systems.",
    "attribution": "Liwei Cheng, Shibo Feng, Lunjie Zhou, Yixuan Guan, Dayan Guan, 'MultiTextEdit: Benchmarking Cross-Lingual Degradation in Text-in-Image Editing', arXiv:2605.08163v2 (2026), CC BY 4.0. Changes: SemAvg divided by 5. No endorsement.",
    "scope": "Table 3 SemAvg column for nine catalogue editing systems (Step1X-Edit rows not used; Bagel-thinking used for BAGEL).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.08163v2",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-whatif-world-ccby",
    "name": "What-If World: A Causal Benchmark for General World Models in Embodied Scenarios (arXiv:2605.27589v1), Table 2",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.27589v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.27589v1",
    "quote": "arXiv abs page for 2605.27589v1 links http://creativecommons.org/licenses/by/4.0/; HTML header 'License: CC BY 4.0'. Sec. 4: 'We evaluate nine video generation models spanning both open-source and closed-source tiers ... All models receive the same initial frame x0 and prompt pair'; App. A.2: 'For every open-source model we used the publicly released default inference configuration' and 'The five closed-source generation models ... were accessed through hosted APIs via kie.ai, and each pair was scored by Gemini 3.1 Pro as a VLM judge'. Authors (UCLA, University of Tennessee) built none of the evaluated models.",
    "attribution": "Kunlin Cai, Rui Song, Jinghuai Zhang, Kaiyuan Zhang, Pranav Bodapati, Alicia Yu, Fnu Suya, Mohammad Rostami, Jiaqi Ma, Yuan Tian, 'What-If World: A Causal Benchmark for General World Models in Embodied Scenarios', arXiv:2605.27589v1 (2026), CC BY 4.0. Changes: pAvg divided by 100. No endorsement.",
    "scope": "Table 2 pAvg column for six catalogue video models (CogVideoX1.5, Cosmos-Predict2 and Kling 3.0 rows not used; see benchmark caveats).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.27589v1",
    "icon": "arxiv.org"
  },
  {
    "id": "visual-cronos-ccby",
    "name": "CRONOS: Benchmarking Counterfactual Physical Consistency in Video Models (arXiv:2605.23699v1), Table 1",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.23699v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.23699v1",
    "quote": "arXiv abs page for 2605.23699v1 links http://creativecommons.org/licenses/by/4.0/; HTML header 'License: CC BY 4.0'. Sec. 4.1: 'Our evaluation includes several state-of-the-art open-source video generation models: Cosmos2.5 [1], CogVideoX1.5 [50], MAGI-1 [45], and Wan2.2 [49]. For Cosmos, we evaluate the two available model sizes 2B and 14B ... All models are evaluated in the I2V setting using the first frame.' Authors (Freiburg, MPI Informatics, CISPA) built none of the evaluated models.",
    "attribution": "León Begiristain, Olaf Dünkel, Adam Kortylewski, 'CRONOS: Benchmarking Counterfactual Physical Consistency in Video Models', arXiv:2605.23699v1 (2026), CC BY 4.0. No endorsement.",
    "scope": "Table 1 I2V 'Success Rate' column for Cosmos2.5-2B, Cosmos2.5-14B, MAGI-1-4.5B and Wan2.2-14B (CogVideoX1.5-5B and the V2V block not used).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.23699v1",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-utter-project",
    "name": "UTTER / EuroLLM official model cards (Hugging Face)",
    "kind": "first-party",
    "match": [
      "huggingface.co/utter-project/EuroLLM-1.7B-Instruct/blob/a25c7fa65fc2a644e6270b8940dbe295b51da681/README.md",
      "huggingface.co/utter-project/EuroLLM-9B-Instruct/blob/f7ae2bc3bcbb538c0b93fa6cfbf388d9898b1ace/README.md",
      "huggingface.co/utter-project/EuroLLM-22B-Instruct-2512/blob/8c6ed5c8d18527d9daed994f645032b86c53f0bb/README.md"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/utter-project/EuroLLM-1.7B-Instruct/blob/a25c7fa65fc2a644e6270b8940dbe295b51da681/README.md",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co",
    "attribution": "UTTER project (EuroLLM) model cards on Hugging Face, observed 2026-10-10",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. (Cards are Apache-2.0; only the sequence length is used.)"
  },
  {
    "id": "first-party-swiss-ai",
    "name": "Swiss AI (Apertus) official model cards (Hugging Face)",
    "kind": "first-party",
    "match": [
      "huggingface.co/swiss-ai/Apertus-8B-Instruct-2509/blob/b946d40447b2b597999b9c86d44bee0b452c919f/README.md",
      "huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/blob/0f2767662a013c0fb631079cea883155252bc3b6/README.md",
      "huggingface.co/swiss-ai/Apertus-v1.5-8B/blob/a411d838600baf0e3635a3daf66fb7c55fc97bb6/README.md",
      "huggingface.co/swiss-ai/Apertus-v1.5-70B/blob/59e744e313e967811aefabde3732609b64201acc/README.md"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509/blob/b946d40447b2b597999b9c86d44bee0b452c919f/README.md",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co",
    "attribution": "Swiss AI Initiative Apertus model cards on Hugging Face, observed 2026-10-10",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped. (Cards are Apache-2.0; only the context length is used.)"
  },
  {
    "id": "card3-ibm-granite-granite-4-0-h-1b",
    "name": "ibm-granite/granite-4.0-h-1b model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ibm-granite/granite-4.0-h-1b/blob/d18cca4c121edb87d022116d281ce212c9136f57/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/ibm-granite/granite-4.0-h-1b/blob/d18cca4c121edb87d022116d281ce212c9136f57/README.md",
    "terms_url": "https://huggingface.co/ibm-granite/granite-4.0-h-1b/blob/d18cca4c121edb87d022116d281ce212c9136f57/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0' (repo tag license:apache-2.0); same Apache-2.0 grant as the other ibm-granite cards in the ledger (card2-ibm-granite-*). Own-lab cells only: the Evaluation Results table prints only IBM's Granite 4.0 Nano models (350M, H 350M, 1B, H 1B).",
    "attribution": "ibm-granite. ibm-granite/granite-4.0-h-1b model card. Source: https://huggingface.co/ibm-granite/granite-4.0-h-1b/blob/d18cca4c121edb87d022116d281ce212c9136f57/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Evaluation Results table (Granite 4.0 Nano family columns) and the architecture table (sequence length).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-ibm-granite-granite-3-1-3b-a800m-instruct",
    "name": "ibm-granite/granite-3.1-3b-a800m-instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ibm-granite/granite-3.1-3b-a800m-instruct/blob/a02780686e08a03fe0d2679a293b5c74a90efa89/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/ibm-granite/granite-3.1-3b-a800m-instruct/blob/a02780686e08a03fe0d2679a293b5c74a90efa89/README.md",
    "terms_url": "https://huggingface.co/ibm-granite/granite-3.1-3b-a800m-instruct/blob/a02780686e08a03fe0d2679a293b5c74a90efa89/README.md",
    "quote": "README front-matter at the pinned revision declares 'license: apache-2.0' and the card says 'License: Apache 2.0'. Own-lab cells only: both Evaluation Results tables list only IBM's Granite-3.1 instruct models (8B, 2B, 3B-A800M, 1B-A400M).",
    "attribution": "ibm-granite. ibm-granite/granite-3.1-3b-a800m-instruct model card. Source: https://huggingface.co/ibm-granite/granite-3.1-3b-a800m-instruct/blob/a02780686e08a03fe0d2679a293b5c74a90efa89/README.md. Licence: Apache-2.0. Changes: selected own-lab numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Open LLM Leaderboard V1 table (ARC-Challenge, HellaSwag, MMLU, WinoGrande, GSM8K) and the V2 table's IFEval column; architecture table (sequence length).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-google-diffusiongemma-26b-a4b-it",
    "name": "google/diffusiongemma-26B-A4B-it model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/README.md",
    "terms_url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0'. Own-lab cells only: the Benchmark Results table prints Google's DiffusionGemma 26B A4B and Gemma 4 26B A4B (both Google models; 'Evaluation results marked in the table are for instruction-tuned models').",
    "attribution": "google. google/diffusiongemma-26B-A4B-it model card. Source: https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/README.md. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Benchmark Results table (both Google columns).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-xhtoken-spark-x2-5-4b",
    "name": "XHToken/Spark-X2.5-4B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/XHToken/Spark-X2.5-4B/blob/0bcb35678590218655dff3765b9e61c83b35e9c4/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/XHToken/Spark-X2.5-4B/blob/0bcb35678590218655dff3765b9e61c83b35e9c4/README.md",
    "terms_url": "https://huggingface.co/XHToken/Spark-X2.5-4B/blob/0bcb35678590218655dff3765b9e61c83b35e9c4/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0'. Evaluation statement: 'We evaluate our models and compare them with leading on-device models of similar size' and the table note '* denotes reported results from publicly-released model cards / papers' (so unstarred comparator cells are the authors' own runs; starred cells are not used). 'All evaluations are conducted in thinking mode.'",
    "attribution": "XHToken. XHToken/Spark-X2.5-4B model card. Source: https://huggingface.co/XHToken/Spark-X2.5-4B/blob/0bcb35678590218655dff3765b9e61c83b35e9c4/README.md. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Benchmark table: Spark-X2.5-4B/1.7B and the unstarred cells of the Qwen3.5 and Gemma 4 comparator columns.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-ai-sage-gigachat3-5-432b-a28b-reasoning",
    "name": "ai-sage/GigaChat3.5-432B-A28B-Reasoning model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/ai-sage/GigaChat3.5-432B-A28B-Reasoning/blob/02f55379ac0b20abd9378791bffd6254725e6f5c/README.md"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/ai-sage/GigaChat3.5-432B-A28B-Reasoning/blob/02f55379ac0b20abd9378791bffd6254725e6f5c/README.md",
    "terms_url": "https://huggingface.co/ai-sage/GigaChat3.5-432B-A28B-Reasoning/blob/02f55379ac0b20abd9378791bffd6254725e6f5c/README.md",
    "quote": "README front matter at the pinned revision declares 'license: mit'. The Benchmark scores table prints GigaChat 3.5 Ultra Instruct, GigaChat 3.5 Ultra Reasoning and DeepSeek V4 Flash Preview Reasoning; the notes give one methodology for all columns (judges, scorers, mini-swe-agent with a three-hour timeout, 'Benchmarks without a methodology-defined system prompt were evaluated with an empty system prompt') and the token-count table compares the two models' own generations, i.e. Sber ran the DeepSeek column.",
    "attribution": "ai-sage. ai-sage/GigaChat3.5-432B-A28B-Reasoning model card. Source: https://huggingface.co/ai-sage/GigaChat3.5-432B-A28B-Reasoning/blob/02f55379ac0b20abd9378791bffd6254725e6f5c/README.md. Licence: MIT. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Benchmark scores table, all three columns (DeepSeek column flagged for owner review).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2608.30181v1-ax-k2",
    "name": "A.X K2 Technical Report (SK Telecom), arXiv 2608.30181v1",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2608.30181v1",
      "arxiv.org/abs/2608.30181v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2608.30181v1",
    "quote": "arXiv abs page for 2608.30181v1 links http://creativecommons.org/licenses/by/4.0/ (checked 2026-10-10). Section 5.1: 'We evaluate A.X K2—alongside its predecessor A.X K1 and leading open-weight baselines—on representative public benchmarks ... Unless otherwise noted, all models are evaluated in thinking mode ... Benchmarks marked with an asterisk (*) are evaluated by Artificial Analysis.' Table 6 caption: 'For benchmarks marked with *, the scores of the other open-weight models are taken from Artificial Analysis.' Only unstarred rows are used.",
    "attribution": "SK Telecom A.X team, A.X K2 Technical Report, arXiv 2608.30181v1 (CC BY 4.0). Changes: selected cells of Tables 6 and 20 transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Table 6 unstarred rows (AIME26, LiveCodeBench v6 Feb-May 2025, SciCode, HLE, GPQA Diamond, IFBench) and Table 20 Global-MMLU-Lite average; A.X K2, A.X K1 (lab) and the open-weight baselines (independent; DeepSeek-V4 Flash column not used: snapshot ambiguous).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-skt",
    "name": "SK Telecom (A.X) official model cards (Hugging Face)",
    "kind": "first-party",
    "match": [
      "huggingface.co/skt/A.X-K2/blob/2287ca456927eed33b899c404c0b2ffaf17aa09f/README.md"
    ],
    "exact": true,
    "license": "Factual first-party specification (not licensed content)",
    "terms_url": "https://huggingface.co/skt/A.X-K2/blob/2287ca456927eed33b899c404c0b2ffaf17aa09f/README.md",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co",
    "attribution": "SK Telecom A.X model cards on Hugging Face, observed 2026-10-10",
    "quote": "Owner policy 2026-09-25: the vendor's own published list price/context limit for its own service, recorded manually as a fact with link and observation date; no page text or compilation reproduced; not scraped."
  },
  {
    "id": "card3-qwen-qwen-agentworld-35b-a3b",
    "name": "Qwen/Qwen-AgentWorld-35B-A3B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/Qwen/Qwen-AgentWorld-35B-A3B/blob/60d2b0434a53d2e62a7c00a489586815d94ebffb/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/Qwen/Qwen-AgentWorld-35B-A3B/blob/60d2b0434a53d2e62a7c00a489586815d94ebffb/README.md",
    "terms_url": "https://huggingface.co/Qwen/Qwen-AgentWorld-35B-A3B/blob/60d2b0434a53d2e62a7c00a489586815d94ebffb/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0'. AgentWorldBench is introduced by the same authors in this release ('Five-dimensional rubric mean per domain, normalized to 0-100 scale'), so every row of the table is the Qwen team's own evaluation; there is no other published source for these numbers.",
    "attribution": "Qwen. Qwen/Qwen-AgentWorld-35B-A3B model card. Source: https://huggingface.co/Qwen/Qwen-AgentWorld-35B-A3B/blob/60d2b0434a53d2e62a7c00a489586815d94ebffb/README.md. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "AgentWorldBench (Open-Ended Evaluation) Overall column for catalogue models.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-jetbrains-mellum2-12b-a2-5b-instruct",
    "name": "JetBrains/Mellum2-12B-A2.5B-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/JetBrains/Mellum2-12B-A2.5B-Instruct/blob/273d9c1c6882ac8e1562e5b7e043857d35a787b6/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Instruct/blob/273d9c1c6882ac8e1562e5b7e043857d35a787b6/README.md",
    "terms_url": "https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Instruct/blob/273d9c1c6882ac8e1562e5b7e043857d35a787b6/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0'. Evaluation statement: 'Post-training evaluation for the instruct (no-thinking) variants ... All values self-reported by JetBrains.' (JetBrains' own runs of every column).",
    "attribution": "JetBrains. JetBrains/Mellum2-12B-A2.5B-Instruct model card. Source: https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Instruct/blob/273d9c1c6882ac8e1562e5b7e043857d35a787b6/README.md. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Post-training evaluation table: Mellum2 column and catalogued comparator columns.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-jetbrains-mellum2-12b-a2-5b-thinking",
    "name": "JetBrains/Mellum2-12B-A2.5B-Thinking model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/JetBrains/Mellum2-12B-A2.5B-Thinking/blob/a7311550557e93cc706ab5dd3d879c1a11703ab4/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Thinking/blob/a7311550557e93cc706ab5dd3d879c1a11703ab4/README.md",
    "terms_url": "https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Thinking/blob/a7311550557e93cc706ab5dd3d879c1a11703ab4/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0'. Evaluation statement: 'Post-training evaluation for the thinking/reasoning variants ... All values self-reported by JetBrains.' (JetBrains' own runs of every column).",
    "attribution": "JetBrains. JetBrains/Mellum2-12B-A2.5B-Thinking model card. Source: https://huggingface.co/JetBrains/Mellum2-12B-A2.5B-Thinking/blob/a7311550557e93cc706ab5dd3d879c1a11703ab4/README.md. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Post-training evaluation table: Mellum2 column and catalogued comparator columns.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-jetbrains-mellum2-1-12b-a2-5b-thinking",
    "name": "JetBrains/Mellum2.1-12B-A2.5B-Thinking model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking/blob/92ddae9fc7665e9f801d141d2e5a6b2caf2460c4/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking/blob/92ddae9fc7665e9f801d141d2e5a6b2caf2460c4/README.md",
    "terms_url": "https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking/blob/92ddae9fc7665e9f801d141d2e5a6b2caf2460c4/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0'. Evaluation statement: 'All models were evaluated by JetBrains with the same pipeline in thinking mode. All values are self-reported by JetBrains.' ('Mellum2 Thinking was re-evaluated with this pipeline').",
    "attribution": "JetBrains. JetBrains/Mellum2.1-12B-A2.5B-Thinking model card. Source: https://huggingface.co/JetBrains/Mellum2.1-12B-A2.5B-Thinking/blob/92ddae9fc7665e9f801d141d2e5a6b2caf2460c4/README.md. Licence: Apache-2.0. Changes: selected numerical cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Evaluation table: Mellum2.1 Thinking, Mellum2 Thinking (lab) and the Gemma 4 E4B / Qwen3.5 9B columns (independent).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2601.09012v2-translategemma",
    "name": "TranslateGemma Technical Report (Google), arXiv 2601.09012v2",
    "kind": "paper",
    "match": [
      "arxiv.org/pdf/2601.09012v2",
      "arxiv.org/abs/2601.09012v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "http://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2601.09012v2",
    "quote": "arXiv abs page for 2601.09012v2 links http://creativecommons.org/licenses/by/4.0/ (checked 2026-10-10). Table 1 compares Google's own TranslateGemma and baseline Gemma 3 models (27B/12B/4B), evaluated by the authors ('We evaluate TranslateGemma using MetricX 24 and Comet22').",
    "attribution": "Google Translate Research Team, TranslateGemma Technical Report, arXiv 2601.09012v2 (CC BY 4.0). Changes: Table 1 Comet22 cells transcribed into JSON; divided by 100. No endorsement.",
    "scope": "Table 1 Comet22 (C22) column for TranslateGemma and Gemma 3 at 4B, 12B and 27B. MetricX (lower-is-better) not recorded: no matching benchmark id.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "arxiv.org"
  },
  {
    "id": "card3-open-thoughts-openthinker2-7b",
    "name": "open-thoughts/OpenThinker2-7B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/open-thoughts/OpenThinker2-7B/blob/57922a00cca8bae9776277bc51c3aceedacc643c/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/open-thoughts/OpenThinker2-7B/blob/57922a00cca8bae9776277bc51c3aceedacc643c/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0' (same as the ledger's card3-open-thoughts-openthinker2-32b). Evaluation statement: 'The numbers reported in the table below are evaluated with our open-source tool Evalchemy.' (covers every row of the table, so comparator rows are the authors' own runs).",
    "attribution": "Open Thoughts team. open-thoughts/OpenThinker2-7B model card. Source: https://huggingface.co/open-thoughts/OpenThinker2-7B/blob/57922a00cca8bae9776277bc51c3aceedacc643c/README.md. Licence: Apache-2.0. Changes: selected numerical cells (own models and comparators evaluated by the authors with Evalchemy) transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Results table cells for catalogue models only.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-open-thoughts-openthinker-7b",
    "name": "open-thoughts/OpenThinker-7B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/open-thoughts/OpenThinker-7B/blob/4ba9dee9b7e6e4796c6bccf6e15c2f5b60843e1e/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/open-thoughts/OpenThinker-7B/blob/4ba9dee9b7e6e4796c6bccf6e15c2f5b60843e1e/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0' (same as the ledger's card3-open-thoughts-openthinker2-32b). Evaluation statement: 'The numbers reported in the table below are evaluated with our open-source tool Evalchemy.' (covers every row of the table, so comparator rows are the authors' own runs).",
    "attribution": "Open Thoughts team. open-thoughts/OpenThinker-7B model card. Source: https://huggingface.co/open-thoughts/OpenThinker-7B/blob/4ba9dee9b7e6e4796c6bccf6e15c2f5b60843e1e/README.md. Licence: Apache-2.0. Changes: selected numerical cells (own models and comparators evaluated by the authors with Evalchemy) transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Results table cells for catalogue models only.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-open-thoughts-openthinker-32b",
    "name": "open-thoughts/OpenThinker-32B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/open-thoughts/OpenThinker-32B/blob/840749ca114a8f02be56f378d2999ad6d9e42d4f/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/open-thoughts/OpenThinker-32B/blob/840749ca114a8f02be56f378d2999ad6d9e42d4f/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0' (same as the ledger's card3-open-thoughts-openthinker2-32b). Evaluation statement: 'The numbers reported in the table below are evaluated with our open-source tool Evalchemy.' (covers every row of the table, so comparator rows are the authors' own runs).",
    "attribution": "Open Thoughts team. open-thoughts/OpenThinker-32B model card. Source: https://huggingface.co/open-thoughts/OpenThinker-32B/blob/840749ca114a8f02be56f378d2999ad6d9e42d4f/README.md. Licence: Apache-2.0. Changes: selected numerical cells (own models and comparators evaluated by the authors with Evalchemy) transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Results table cells for catalogue models only.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "first-party-nous-research-scores",
    "name": "Nous Research vendor-reported scores for its own Hermes models (Hermes 3 Technical Report)",
    "kind": "first-party",
    "match": [
      "arxiv.org/pdf/2408.11857v1",
      "arxiv.org/abs/2408.11857v1"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://arxiv.org/abs/2408.11857v1",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "arxiv.org",
    "attribution": "Nous Research, Hermes 3 Technical Report (arXiv 2408.11857v1); Hermes 3 8B/70B/405B cells of Table 5 hand-recorded 2026-10-10; percentages converted to fractions. Lab-reported. No endorsement.",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. arXiv licence for v1 is the non-exclusive distribution licence (abs page, checked 2026-10-10), so the Llama 3.1 Instruct comparator columns of Table 5 are not used."
  },
  {
    "id": "card3-huggingfacetb-smollm2-360m-instruct",
    "name": "HuggingFaceTB/SmolLM2-360M-Instruct model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct/blob/a10cc1512eabd3dde888204e902eca88bddb4951/README.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct/blob/a10cc1512eabd3dde888204e902eca88bddb4951/README.md",
    "quote": "README front matter at the pinned revision declares 'license: apache-2.0'. Evaluation statement: 'All evaluations are zero-shot unless stated otherwise, and we use lighteval to run them.' (covers every column, including the Qwen2.5-0.5B-Instruct comparator).",
    "attribution": "Hugging Face TB. HuggingFaceTB/SmolLM2-360M-Instruct model card. Source: https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct/blob/a10cc1512eabd3dde888204e902eca88bddb4951/README.md. Licence: Apache-2.0. Changes: selected cells transcribed into JSON; percentages divided by 100. No endorsement.",
    "scope": "Instruction-model table: IFEval (average prompt/inst), HellaSwag, PIQA, BBH 3-shot, GSM8K 5-shot for SmolLM2-360M-Instruct and Qwen2.5-0.5B-Instruct.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "card3-liquidai-lfm2-1-2b",
    "name": "LiquidAI/LFM2-1.2B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/README.md"
    ],
    "exact": true,
    "license": "LFM Open License v1.0",
    "license_url": "https://huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/LICENSE",
    "terms_url": "https://huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/LICENSE",
    "quote": "Same licence and treatment as the ledger's card-liquidai-lfm2-5-1-2b-thinking: the card front matter links the LFM Open License v1.0 (license_name lfm1.0), which covers documentation and grants reproduction, adaptation, display and distribution subject to its terms (Section 5 commercial revenue / Legal Entity conditions apply to recipients). Own-model cells only.",
    "conditions": "LFM Open License v1.0 includes commercial revenue and Legal Entity conditions. Downstream users must assess their own eligibility; this compilation does not remove these restrictions.",
    "attribution": "Liquid AI. LiquidAI/LFM2-1.2B model card. Source: https://huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/README.md. Licensed under LFM Open License v1.0. Modified for whatmodel.sh: selected own-model cells, percentages to fractions. No endorsement.",
    "scope": "'Automated benchmarks' table: LFM2-350M / 700M / 1.2B rows (by: lab) and, since pass 12, the Qwen3-0.6B, Qwen3-1.7B, Llama-3.2-1B-Instruct, gemma-3-1b-it comparator rows (by: independent, read as Liquid's own runs) for MMLU, GPQA, IFEval, IFBench, GSM8K, MGSM, MMMLU; and the model-property table (context length).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2605.07492v2-puredocbench",
    "name": "How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings (arXiv:2605.07492v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.07492v2",
      "arxiv.org/abs/2605.07492v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.07492v2",
    "quote": "arXiv abs page for 2605.07492v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Abstract: 'We evaluate 58 models, including pipeline-based parsers, end-to-end parsers, and general vision-language models.' Sec. 3.1 'Models (58). We prioritize recent releases and compare 13 pipeline / multi-stage specialists, 19 end-to-end specialists, and 26 general-purpose VLMs (Table 2) ... General VLMs span the Qwen and MiniCPM families alongside Claude Opus 5, Gemini 3.6 Flash, GLM-5.3-Flash, GPT-5.6 Sol, and Kimi K2.6.' Intro: 'We release document images, sources, annotations, generation and evaluation tools, and available model predictions for reproducible evaluation.' The benchmark is new in this paper, so every score is the authors' own run. Authors (CAS and universities) made none of the evaluated general VLMs. Checkpoint links (e.g. Qwen3-VL-8B -> Qwen/Qwen3-VL-8B-Instruct) from the authors' repo README at github.com/zhihengli-casia/PureDocBench commit 154f576b9ef9aabd2460b4276432d81bf5a268c8.",
    "attribution": "Zhiheng Li, Zongyang Ma, Jiaxian Chen et al. (UCAS / CASIA, NWPU, JLU, USTC, PKU, HelloGroup, ShanghaiTech), 'How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings', arXiv:2605.07492v2 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (three-track parsing accuracy, 58 models), final 'Overall' column (mean of the Clean/Digital/Real track Overalls, Avg3) for the 'General-purpose VLMs' block (21 catalogued models). Pipeline and end-to-end OCR specialists, MiniCPM-V-4.5/4.6 and Step3-VL are not catalogued. Per-track and component columns not used (correlated slices).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.07492v2",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2605.08366v1-swe-atlas",
    "name": "SWE Atlas: Benchmarking Coding Agents Beyond Issue Resolution (arXiv:2605.08366v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.08366v1",
      "arxiv.org/abs/2605.08366v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.08366v1",
    "quote": "arXiv abs page for 2605.08366v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 3: 'We run a number of frontier coding models as well as the top open models on the SWE Atlas benchmark suite. We use the Harbor Framework to run reproducible inference and evaluations ... In addition, we use mini-SWE-agent, which uses just the bash command as a tool for the agent, as a minimal harness to evaluate some models under a common scaffold.' Table 1: 'Resolution rates on SWE Atlas, averaged over 3 trials. Pass@1 is the mean per-trial pass rate.' Table 8 gives thinking settings (GPT-5.4 and Opus 4.7 xHigh; Opus/Sonnet 4.6 and Gemini 3.1 Pro High; open models default/High). Authors (Scale AI) made none of the models.",
    "attribution": "Mohit Raghavendra, Soham Dan, Miguel Romero Calvo et al. (Scale AI), 'SWE Atlas: Benchmarking Coding Agents Beyond Issue Resolution', arXiv:2605.08366v1 (2026), CC BY 4.0. Changes: selected cells of Table 1 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 1 'Pass@1' column for the mini-SWE-Agent block (one common scaffold). Native-scaffold rows (Codex, Claude Code, Gemini CLI) are a different harness and not imported. 'Gemini 3 Flash' row not imported (Table 8 calls it 'Gemini 3.1 Flash'; identity ambiguous). Per-workflow columns not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.08366v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2605.08614v1-diagnosticiq",
    "name": "DiagnosticIQ: A Benchmark for LLM-Based Industrial Maintenance Action Recommendation from Symbolic Rules (arXiv:2605.08614v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.08614v1",
      "arxiv.org/abs/2605.08614v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.08614v1",
    "quote": "arXiv abs page for 2605.08614v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 1: 'We benchmark 29 LLMs across 9 families and 4 embedding baselines on DiagnosticIQ and five targeted variants.' Sec. 4: 'We evaluate models via direct zero-shot prompting ... we report both standard accuracy and macro-averaged accuracy' (macro over asset classes). Appendix A.16.2: 'For LLM inference we explicitly specify temperature = 0 and maximum tokens = 4096. All other parameters are left at their defaults.' Core DiagnosticIQ questions have 4 options (Pro: 10). IBM authors made the Granite models (rows marked by: lab); all other rows are independent.",
    "attribution": "Devin Yasith De Silva, Dhaval Patel, Christodoulos Constantinides, Shuxin Lin, Nianjun Zhou, Paul J Adams, Sal Rosato, Nicolas Constantinides, Deborah L. McGuinness, Jayant Kalagnanam (IBM, Rensselaer Polytechnic Institute), 'DiagnosticIQ: A Benchmark for LLM-Based Industrial Maintenance Action Recommendation from Symbolic Rules', arXiv:2605.08614v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (leaderboard), 'Macro % D.IQ' column (core 4-option set, 6,690 questions). Not imported: deepseek-v3, mistral-large and gemini-1.5-pro (no snapshot given; several catalogued), gpt-oss-120b (30.38 macro on the 4-option core set but 32.79 on the 10-option Pro set: consistent with an answer-format failure), and the embedding-only baselines.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.08614v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2605.10912v1-wildclawbench",
    "name": "WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation (arXiv:2605.10912v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.10912v1",
      "arxiv.org/abs/2605.10912v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.10912v1",
    "quote": "arXiv abs page for 2605.10912v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4: 'Models and Harnesses. We evaluate 19 frontier models on WildClawBench under four harnesses: OpenClaw (the default), Claude Code, Codex, and Hermes Agent. All models are shipped through a unified OpenRouter endpoint, and each harness ships as a dedicated Docker image with pinned OS, Python toolchain, and pre-installed binaries.' Table 2 caption: 'Main results on WildClawBench under the OpenClaw harness ... The Overall column is a task-count-weighted average of the Multimodal (26 tasks) and Pure Text (34 tasks) columns.' WildClawBench is introduced in this paper, so the scores are the authors' runs; Shanghai AI Lab's own models are not in Table 2.",
    "attribution": "Shuangrui Ding, Xuanlang Dai, Long Xing et al. (Shanghai AI Laboratory, CUHK, Fudan, USTC, SJTU, Tsinghua, Shanghai Innovation Institute, ZJU, NTU), 'WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation', arXiv:2605.10912v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (main results under the OpenClaw harness), 'Overall Score' column (60 tasks). Other harnesses and the Multimodal/Pure-Text split columns not used. Reuses the existing 'wildclawbench' benchmark id (this paper defines the benchmark). GLM 5 Turbo not catalogued.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.10912v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2605.17281v2-contractbench",
    "name": "ContractBench: Can LLM Agents Preserve Observation Contracts? (arXiv:2605.17281v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2605.17281v2",
      "arxiv.org/abs/2605.17281v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2605.17281v2",
    "quote": "arXiv abs page for 2605.17281v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4: 'We evaluate 38 model variants across 15 families ... Each model is run on the full 33-task ContractBench suite at k=3 rollouts per task (n=99 episodes per model), with temperature 0, a task-describing system prompt, and a 600-second per-episode wall-clock timeout.' App.: 'ContractBench evaluation runs against two backend types: external LLM APIs (proprietary frontier models and many open-source models hosted by inference providers) and a local vLLM server (for the smaller open-source variants we hosted ourselves).' Table 9 pins every model identifier. Authors (UC Davis / USC / HKU) built none of the evaluated models.",
    "attribution": "Jicheng Wang, Yifeng He, Zili Wang, Hanwen Xing, Arkaprava De, Hao Chen, 'ContractBench: Can LLM Agents Preserve Observation Contracts?', arXiv:2605.17281v2 (2026), CC BY 4.0. Changes: selected cells of Table 10 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 10 master leaderboard, SR% column, frontier-proprietary and open-source-SOTA blocks (24 catalogued models; identifiers from Table 9). The 'Sub-cliff' block (all rows 0.0%, mostly Base checkpoints with no chat template/tool-call format, plus DeepSeek-R1 via Together and Qwen2.5 Instruct via Featherless) is not imported: an all-zero tool-calling row cannot be told apart from a tool-format failure.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2605.17281v2",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.01348v3-chartarena",
    "name": "ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats (arXiv:2606.01348v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.01348v3",
      "arxiv.org/abs/2606.01348v3"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.01348v3",
    "quote": "arXiv abs page for 2606.01348v3 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 5.1: 'We evaluate 26 representative models across three categories: (a) General-purpose MLLMs (16 models) ... (Qwen2.5-VL, InternVL3.5, Qwen3-VL, GLM-4.5V, Qwen3.5-35B-A3B, Kimi K2.5) and proprietary models (GPT-4o, GPT-5, Gemini 2.5/3.1 Pro, Seed-1.8/2.0, MiMo-V2-Omni) ... Each model is run in its native output format, and outputs are normalized before scoring ... for general-purpose MLLMs we adopt a unified prompting template carefully tuned for chart parsing. All models are evaluated under identical inference settings.' Table C.3 gives 'additional results across various model groups and parameter scales' under the same protocol. ChartArena is new in this paper, so all scores are the authors' runs. The only Tencent model in the tables (HunyuanOCR) is not catalogued, so all imported rows are independent.",
    "attribution": "Shangpin Peng, Gengluo Li, Xingyu Wan, Chengquan Zhang, Hao Feng, Binghong Wu, Huawen Shen, Weinong Wang, Ziyi Cai, Zhuotao Tian, Han Hu, Can Ma, Yu Zhou (Tencent LLM Department, SLAI, HKUST, CAS IIE, Nankai), 'ChartArena: Benchmarking Chart Parsing across Languages, Scenarios, and Formats', arXiv:2606.01348v3 (2026), CC BY 4.0. Changes: selected cells of Tables 2 and C.3 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (main results) and Table C.3 (additional results, same protocol), English 'Average' column (mAP_high averaged over the eight chart families and three visual scenarios). The Chinese (ZH) average is not imported (one headline per model). GPT-4o (no snapshot), InternVL3.5, Seed-1.8/2.0, MiniCPM-V 4.5, document-parsing and chart-expert models are not catalogued or ambiguous; Molmo-7B-D/72B rows (1.6/2.1 mAP, near zero on every chart type) not imported as likely output-format failures.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2606.01348v3",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.02404v1-k-browsecomp",
    "name": "K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts (arXiv:2606.02404v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.02404v1",
      "arxiv.org/abs/2606.02404v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.02404v1",
    "quote": "arXiv abs page for 2606.02404v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 5: 'We evaluate proprietary models and open-weight models on K-BrowseComp-Verified. The proprietary baselines include GPT-5.5, GPT-5.4-mini, and Gemini-3.1-Flash-Lite. The open-weight baselines include DeepSeek-V4-Pro, GLM-5.1, Qwen3.6-35B-A3B, and Gemma-4-31B-it. We also evaluate Korean open-weight models, including K-EXAONE-236B-A23B, A.X-4.0, HyperCLOVAX-SEED-Think-32B, and Kanana-2-30B-A3B-Thinking-2601.' Evaluation protocol: 'We build our browsing-agent evaluation on the search_evals framework ... deep-research agent and ... Perplexity Search as the search backend ... a budget of 10 search calls per question ... Each model is evaluated once on each of the 300 questions in the verified subset. We use GPT-5.4-mini to extract the final answer ... We report the single-run accuracy, which corresponds to pass@1.' One co-author affiliation is NAVER Cloud AI, so the HyperCLOVA X SEED row is by: lab; all others independent.",
    "attribution": "Nahyun Lee, Dongkeun Yoon, Guijin Son et al. (Chung-Ang University, KAIST, Seoul National University, OnelineAI, NAVER Cloud AI, Carnegie Mellon University), 'K-BrowseComp: A Web Browsing Agent Benchmark Grounded in Korean Contexts', arXiv:2606.02404v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 'Verified subset Pass@1 Acc.' column (300 human-verified questions). Synthetic-subset column not used. Not imported: A.X-4.0 (not catalogued) and Kanana-2-30B-A3B-Thinking-2601 (not catalogued; 0.00).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2606.02404v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.05104v2-kina",
    "name": "Knowledge Index of Noah's Ark (arXiv:2606.05104v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.05104v2",
      "arxiv.org/abs/2606.05104v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.05104v2",
    "quote": "arXiv abs page for 2606.05104v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 5: 'We evaluate 42 models from 13 AI labs on KINA, spanning closed-source flagship APIs, open-source dense and Mixture-of-Experts (MoE) checkpoints, and reasoning models. The full model list is in Appendix F.1. All results are reported as avg@4 accuracy at default temperature, with 32K maximum new tokens and a reasoning budget of 16K (or \"Medium\"); prompts are in Appendix F.2.' Authors' affiliations (2077AI, M-A-P community, universities) did not make any evaluated model.",
    "attribution": "Sheng Jin, Minghao Liu, Yunze Xiao et al. (2077AI, M-A-P, CMU, Brown, Waseda, UTokyo, MIT, Arizona, Northwestern, Duke-NUS), 'Knowledge Index of Noah's Ark', arXiv:2606.05104v2 (2026), CC BY 4.0. Changes: selected cells of Table 3 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 3 (full leaderboard of 42 models), ALL column only (per-discipline columns are correlated slices); identities from Table 13 (Appendix F.1). Doubao-Seed-2.0-Pro/Lite-260215 not catalogued. Tool-use (web search) Table 6 not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2606.05104v2",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.08071v1-surgiq",
    "name": "SurgiQ: A Large-Scale Multi-Domain Benchmark for Evaluating Surgical Understanding in Large Language Models (arXiv:2606.08071v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.08071v1",
      "arxiv.org/abs/2606.08071v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.08071v1",
    "quote": "arXiv abs page for 2606.08071v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4: 'We evaluate SurgiQ in a zero-shot multiple-choice setting using a diverse set of open-weight large language models, including general-purpose, biomedical, and reasoning-oriented architectures.' Sec. 4.2: 'We evaluate all models under a unified zero-shot multiple-choice setting ... For open-weight models, predictions are computed from next-token log-probabilities over the answer options ... answer choices are randomly shuffled once during dataset construction and the resulting order is kept fixed across all evaluations ... the evaluation procedure is fully deterministic.' Table A6 lists the Hugging Face repositories. Authors (MBZUAI) made none of the imported models (MBZUAI's K2-Think-V2 and MedMO rows are not catalogued and not imported).",
    "attribution": "Ayah Al-Naji, Edoardo Fazzari, Saif Alkindi, Hamdan Alhadhrami, Preslav Nakov, Cesare Stefanini (MBZUAI), 'SurgiQ: A Large-Scale Multi-Domain Benchmark for Evaluating Surgical Understanding in Large Language Models', arXiv:2606.08071v1 (2026), CC BY 4.0. Changes: selected cells of Table 4 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 4 'Avg.' column (micro-accuracy over all SurgiQ items after duplicate filtering) for catalogued instruction/chat checkpoints and the two Meditron base models; repositories per Table A6. Not imported: Qwen2.5-32B/72B base rows, MedGemma-27B-IT (text vs multimodal repo not stated), TxGemma, Meerkat, MedMO, K2-Think-V2, OpenBioLLM, Med42 (not catalogued).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2606.08071v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.11070v1-t1-bench",
    "name": "T1-Bench: Benchmarking Multi-Scenario Agents in Real-World Domains (arXiv:2606.11070v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.11070v1",
      "arxiv.org/abs/2606.11070v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.11070v1",
    "quote": "arXiv abs page for 2606.11070v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 3.1: 'We evaluate six open-weight models, including GPT-OSS 20B, GPT-OSS 120B, Gemma4-26B-A4B-it, Gemma4-31B-it, Gemma4-E2B-it, and Gemma4-E4B-it. In addition, we include six proprietary models: Claude Haiku 4.5, Claude Sonnet 4.5, Claude Opus 4.6, GPT-5.4-Nano, GPT-5.4-Mini, and GPT-5.4. For consistency across evaluation and simulation settings, all user and judge agents are instantiated using Gemma4-26B-A4B-it.' Sec. 3.2: a conversation passes only with perfect tool recall, perfect parameter recall and exact match of all tool outputs; 'For all experiments, we set K=3.' Appendix Table 6 gives per-model sampling and reasoning effort (temperature 1.0). Authors (Capital One) made none of the models.",
    "attribution": "Genta Indra Winata, Amartya Chakraborty, Yuzhen Lin et al. (AI Foundations, Capital One), 'T1-Bench: Benchmarking Multi-Scenario Agents in Real-World Domains', arXiv:2606.11070v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (overall results averaged over all domains), 'Pass^k' column (K=3: probability that all three sampled trials pass). Pass@k, tool/parameter accuracy and LLM-judge columns not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2606.11070v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2606.12250v1-polish-medical-exams",
    "name": "Reassessing High-Performing LLMs on Polish Medical Exams: True Competence or Bias-Driven Performance? (arXiv:2606.12250v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2606.12250v1",
      "arxiv.org/abs/2606.12250v1"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://arxiv.org/abs/2606.12250v1",
    "quote": "arXiv abs page for 2606.12250v1 links http://creativecommons.org/licenses/by-sa/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY-SA 4.0'. Abstract: 'We evaluate 21 LLMs and show that evaluation design strongly affects results.' Sec. 6: 'the LLMs are evaluated on the unmodified exam sets, comparing two evaluation techniques: the technique reproduced from the Polish Medical Exams benchmark Grzybowski et al. (2025) and the introduced one that changes the instruction to minimize bias and utilizes constrained generation.' Sec. 6.1 names the checkpoints (Bielik-11B-v3-Instruct, Bielik-4.5B-v3.0-Instruct, PLLuM-12B-instruct, Llama-PLLuM-70B-instruct, Qwen3.5 9B/35B/122B, Llama3.3-70B, Gemma3 12B/27B, BioMistral-7B, Llama3-OpenBioLLM-70B, MedGemma 4B/27B ...). Appendix: 'the generation of the answers was done with a temperature equal to 0, constrained generation (when applicable) was limited to answers A, B, C, D, E' using vLLM/Transformers on the authors' A100 cluster. The authors made none of the evaluated models.",
    "attribution": "Antoni Lasik, Jakub Pokrywka, Łukasz Grzybowski, Jeremi Ignacy Kaczmarek, Gabriela Korzańska, Janusz Świeczkowski-Feiz, Oskar Pastuszek, Paulina Hoffman, Jakub Tomasz Dąbrowski, Wojciech Kusa, 'Reassessing High-Performing LLMs on Polish Medical Exams: True Competence or Bias-Driven Performance?', arXiv:2606.12250v1 (2026), CC BY-SA 4.0. Changes: selected cells of Table 4 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 4 'Avg OM' column (accuracy under the authors' bias-mitigated protocol, averaged over the LEK, LDEK, PES, PESDL, PESF, LEK-EN and LDEK-EN exams). 'Avg PM' (replicated earlier protocol) and the open-ended/modified-question tables are not used. Not imported: PLLuM-12B-NC-Inst/Chat (2412 vs 2507 release not stated), LLaMA3-8B (base vs Instruct not stated), MedGemma-27B (text vs multimodal), MediPhi, BioMistral-7B-DARE, Meditron3-70B (not catalogued).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2606.12250v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2608.10408v1-viseditbench",
    "name": "VisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback? (arXiv:2608.10408v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2608.10408v1",
      "arxiv.org/abs/2608.10408v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2608.10408v1",
    "quote": "arXiv abs page for 2608.10408v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 1: 'We evaluate 20 state-of-the-art VLMs on VisEditBench in a zero-shot setting using executability, task accuracy, readability and clarity, visual quality, visual similarity, and final pass rate.' Sec. 4: 'We use a fixed decoding configuration for all models to support reproducibility.' Table 10: 'image detail = high, temperature = 0.2, top-p = 1.0, maximum output tokens = 4096 ... GPT-4o is used as the default evaluator; for GPT-4o outputs, Gemini 2.5 Pro is used to avoid self-evaluation bias.' Strict pass: executable AND task accuracy >= 4.5 AND readability >= 4.0 AND visual quality >= 4.0 AND visual similarity >= 90. Authors (York University et al.) made none of the evaluated models.",
    "attribution": "Mizanur Rahman, Arshia Azimlu, Shadikur Rahman, Md Tahmid Rahman Laskar, Amran Bhuiyan, Shafiq Joty, Enamul Hoque Prince (York University et al.), 'VisEditBench: Can Vision-Language Models Edit Visualization Code from Multimodal Feedback?', arXiv:2608.10408v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 'Overall' pass-rate column for catalogued models with unambiguous identity. Not imported: GPT-4o (no snapshot), Qwen3-VL-2B/4B/8B/30B-A3B/32B (Instruct vs Thinking not stated), InternVL-3.5 (not catalogued), and Gemini-3.0-Flash (17.06%: only 46.02% of its outputs executed vs 82-96% for every other model, Table 3, consistent with the 4,096-token output cap rather than editing ability).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2608.10408v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2608.12262v1-diagram-mmu",
    "name": "Diagram-MMU: A Multi-Modal Benchmark for Scientific Diagrams (arXiv:2608.12262v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2608.12262v1",
      "arxiv.org/abs/2608.12262v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2608.12262v1",
    "quote": "arXiv abs page for 2608.12262v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4: 'We evaluate 12 closed-source and open-source MLLMs on Diagram-MMU.' Table D.1 'Generation configurations of evaluated models' (API or vLLM, temperature 1.0, max 16,384 output tokens for the imported models; GPT-5.2 reasoning_effort = medium). DQA scoring: 'we evaluate diagram question answering using accuracy. We use Qwen3-Next-80B-A3B-Instruct to extract the answer and assign binary scores' (judge agreement with human majority 97.5%, Table C.2). The paper is new, so all scores are the authors' runs; Baidu co-authors, but no Baidu model is evaluated.",
    "attribution": "Weihao Bo, Shan Zhang, Yanpeng Sun, Jie Liu, Yongke Yao, Jinhao Du, Wei He, Kai Zou, Zechao Li, Jingdong Wang (NJUST, Baidu, Adelaide, SUTD, Southeast U., ECNU, Oxford, NetMind.ai), 'Diagram-MMU: A Multi-Modal Benchmark for Scientific Diagrams', arXiv:2608.12262v1 (2026), CC BY 4.0. Changes: selected cells of Table 5 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 5 (main results on foundational capability, full dataset), 'DQA Acc.' column only (the D2C parsing/editing columns are code-similarity metrics). Not imported: Qwen3-VL-235B-A22B and Qwen3-VL-8B (Instruct vs Thinking not stated), Seed-2.0 Pro, InternVL3-38B, TikZero+ (not catalogued). The agentic settings (Tables 6-7) use a 300-diagram mini split and are not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2608.12262v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2608.17895v1-bear-bench",
    "name": "BEAR-Bench: A Bilingual Enterprise and Academic Reasoning Benchmark for Multimodal Models (arXiv:2608.17895v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2608.17895v1",
      "arxiv.org/abs/2608.17895v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2608.17895v1",
    "quote": "arXiv abs page for 2608.17895v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 5.1: 'We evaluate a diverse set of MLLMs on BEAR-Bench, covering both open-weight models (Qwen3.5-0.8B/4B/9B/27B, Qwen3-VL-2B/8B-Instruct, Qwen3-VL-4B-Thinking, and Gemma-4-31B-it) and proprietary systems (Qwen3.5-397B-A17B, Qwen 3.6 Plus, Gemini 2.5/3.1 Pro/Flash, and Claude Sonnet/Opus 4.6). The open-weight models were run locally on an internal GPU server equipped with NVIDIA H100 and NVIDIA L40 accelerators; the proprietary ones were queried via the OpenRouter API. Inference protocol. All models were evaluated zero-shot under a fixed protocol ... a uniform decoding temperature of 0.6.' Scoring: GPT-4o LLM-as-a-judge binary semantic-equivalence verdict (validated against a 200-verdict human audit). No evaluated model is from the authors' organisations.",
    "attribution": "Liubov Chubarova, Alexandra Kuleshova, Daniil Volkov, Kirill Sultanov, Alexey Zaytsev (Yandex, Applied AI Institute), 'BEAR-Bench: A Bilingual Enterprise and Academic Reasoning Benchmark for Multimodal Models', arXiv:2608.17895v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (BEAR-Bench leaderboard), 'Overall' column (n=1000). Subset columns not used. 'Gemini 3.1 Flash' row not imported (no Google model of that exact name; ambiguous between Gemini 3 Flash and Gemini 3.1 Flash-Lite).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2608.17895v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2608.19741v4-thinkingbox",
    "name": "One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business Workflows (arXiv:2608.19741v4)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2608.19741v4",
      "arxiv.org/abs/2608.19741v4"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2608.19741v4",
    "quote": "arXiv abs page for 2608.19741v4 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 5: 'We evaluate a diverse set of proprietary frontier models and open-weight agentic models ... Each model is evaluated in the same Thinkingbox sandbox with the same system prompts, tool definitions, simulated-user behavior, and task-specific executable checks ... we run each model on each task for N independent trials ... A trial is counted as correct only when all executable checks for the task pass.' Appendix: API models 'are accessed through Azure-hosted API deployments' (temperature 1.0, reasoning_effort medium, Table 9); Qwen models served locally with vLLM (Table 10); simulated user and response judge are a fixed GPT-5.4-mini. Microsoft co-authors made none of the evaluated models.",
    "attribution": "Zhuochun Li, Youngmin Ko, Ali Keramati, Tuhin Kundu et al. (Microsoft, University of Pittsburgh, Northwestern, UC Irvine, Columbia), 'One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business Workflows', arXiv:2608.19741v4 (2026), CC BY 4.0. Changes: selected cells of Table 15 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 15 'pass@1' column (micro-averaged over 507 tasks x N=20 trials; equals the Table 4 'Average' for the models in both tables) for the base models; the authors' RL fine-tuned Qwen variants are not catalogued checkpoints. Not imported: DeepSeek-V4-Pro (Azure deployment; April preview vs 0813 release not stated) and MiniMax-M2.5 (0.19%, 0/20 passes on 504 of 507 tasks: indistinguishable from a serving/format failure).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2608.19741v4",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2609.05141v2-scidocbench",
    "name": "SciDocBench: A Workflow-Centered Benchmark and Data Pipeline for Scientific Document Understanding (arXiv:2609.05141v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.05141v2",
      "arxiv.org/abs/2609.05141v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.05141v2",
    "quote": "arXiv abs page for 2609.05141v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 5: 'We evaluate 13 proprietary and open multimodal models on all 496 matched SciDocBench instances, reporting overall, capability-group, and setting-level scores.' Appendix F.1: 'The leaderboard evaluates 13 proprietary and open multimodal models on the same 124 underlying questions under the four settings ... yielding 496 instances per model. The input contains only the question and its source materials; reference answers, task labels, and evaluator metadata are withheld. Overall scores average all instances ... Failed or unusable responses receive zero.' Scoring: deterministic rules (57 questions), a rubric-based semantic LLM judge (66; audited against humans, Table 16) and execution (1). Shanghai AI Lab co-authors' models (InternVL) are not in the leaderboard.",
    "attribution": "Shenxi Wu, Yuhong Liu, Haosong Zhang et al. (CUHK, Shanghai AI Laboratory, NYU, Fudan, SJTU, HIT, JD Explore Academy, CPII), 'SciDocBench: A Workflow-Centered Benchmark and Data Pipeline for Scientific Document Understanding', arXiv:2609.05141v2 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (SciDocBench leaderboard), 'Overall' column (all 496 instances). Setting and capability columns not used. Qwen3.8-Max not imported (undated 'Qwen3.8-Max' accessed 2026-09-04, after the 0902 snapshot; catalogue keeps the undated release and the 0902 snapshot apart).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2609.05141v2",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2609.21562v2-gamelogicbench",
    "name": "GameLogicBench: Evaluating Coding Agents on Runtime Game Logic with Tick-Level State Assertions (arXiv:2609.21562v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.21562v2",
      "arxiv.org/abs/2609.21562v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.21562v2",
    "quote": "arXiv abs page for 2609.21562v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4: 'We evaluate twenty configurations, each pairing a model with one of three scaffolds: Claude Code 2.1.177, Codex 0.144.1, or OpenCode 1.17.18. Both solving and judging use Godot 4.4. All solve sessions use effort=high and a 3,600-second time limit.' Table 3 caption: 'Solve rates per tier, one sample per configuration, all runs with egress sealed.' Appendix A: 'For each DeepSeek line, the undated name denotes the original April 2026 release and the dated suffix denotes a newer snapshot.' Table 6 lists the API model ids and release dates. Authors (Nanjing University et al.) made none of the models.",
    "attribution": "Xinyu Che, Yunfei Ge, Shihao Li, Yanchen Liu, Hang Yan, Xinping Lei, Yanghai Wang, Zixuan Dong, Yifan Yao, Qianqian Xie, Letian Zhu, Jiaheng Liu (Nanjing University et al.), 'GameLogicBench: Evaluating Coding Agents on Runtime Game Logic with Tick-Level State Assertions', arXiv:2609.21562v2 (2026), CC BY 4.0. Changes: selected cells of Table 3 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 3 'Solve rate (%) Average' column for the twelve configurations run with the Claude Code 2.1.177 scaffold (one harness per run). Codex and OpenCode scaffold-ablation rows not imported. Per-tier columns not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2609.21562v2",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2609.31632v1-eegagentbench",
    "name": "EEGAgentBench: Benchmarking LLM Agents on Short- and Long-Horizon EEG Analysis (arXiv:2609.31632v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.31632v1",
      "arxiv.org/abs/2609.31632v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.31632v1",
    "quote": "arXiv abs page for 2609.31632v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4.1: 'We evaluate 29 LLMs from 15 model families on EEGAgentBench ... Each model is evaluated independently on all 1,072 instances across the six tasks. For T1, models answer EEG knowledge questions directly. For T2-T6, every model has access to the same set of 10 EEG analysis tools and may iteratively invoke them before submitting a final answer ... results are reported from a single run for each model-instance pair ... The Overall score is the equally weighted mean of T1 accuracy, Macro-F1 for T2-T4 and T6, and Event-F1 for T5.' Each model uses its own context window (Sec. 6.5). Authors (ICT, CAS) made none of the models.",
    "attribution": "Huyu Wu, Weining Weng, Yuchen Liu, Yiqiang Chen, Yang Gu (Institute of Computing Technology, Chinese Academy of Sciences), 'EEGAgentBench: Benchmarking LLM Agents on Short- and Long-Horizon EEG Analysis', arXiv:2609.31632v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 (main results), 'Overall' column. Per-task columns not used. Not imported: DeepSeek-V4-Pro and DeepSeek-V4-Flash (no release/snapshot stated; the Flash price in Table 6 is not DeepSeek's list price, so the served version is unclear) and ERNIE-4.5-VL-424B (not catalogued).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2609.31632v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2609.38335v1-e2e-swe",
    "name": "E2E-SWE: Benchmarking LLMs on Building Working Codebases from Scratch (arXiv:2609.38335v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.38335v1",
      "arxiv.org/abs/2609.38335v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.38335v1",
    "quote": "arXiv abs page for 2609.38335v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 3.1: 'We evaluate thirteen frontier models across six model families: Claude, GPT, Muse Spark, Gemini, Kimi, and GLM. We select the strongest reasoning-effort setting that runs reliably within the 128K per-request output-token limit ... We run each model four times independently over the full corpus and report the average.' Agent scaffold: 'All models are evaluated with a simple ReAct agent, similar to mini-SWE-agent ... Each rollout has a 4-hour wall-clock budget.' Metric: 'pass@1: the fraction of tasks fully resolved by a single rollout, where resolution requires the generated repository to pass every hidden test.' Authors are Meta employees, so the Muse Spark 1.3 row is by: lab; all others independent.",
    "attribution": "Hantian Ding, Chloe Bi, Jiacheng Zhu, John Yang, Matt Deitke, Pengcheng Yin, Zijian Wang, Rui Hou (Meta Superintelligence Labs), 'E2E-SWE: Benchmarking LLMs on Building Working Codebases from Scratch', arXiv:2609.38335v1 (2026), CC BY 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 'Pass@1' column for all 13 models (one shared ReAct scaffold, 4 runs averaged). NL2Repo-Bench appendix results not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2609.38335v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2609.38480v2-klinikebench",
    "name": "KlinikeBench: Evaluating Language Models Beyond Diagnostic Accuracy (arXiv:2609.38480v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2609.38480v2",
      "arxiv.org/abs/2609.38480v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2609.38480v2",
    "quote": "arXiv abs page for 2609.38480v2 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 3.1: 'We integrate KlinikeBench into Harbor ... The evaluated model uses the ReAct-style Terminus-2 harness to question the patient, execute tools through the clinic command interface, and record a diagnosis ... each model receives the same tasks and tool interface, with one rollout per model-task pair.' Sec. 3.2: 'We evaluate 31 models from seven families: Claude, GPT, Gemini, Grok, DeepSeek, Kimi, and GLM (Table 7). Medium reasoning effort is used across all models.' GPT-5.4-mini is the patient simulator and must-ask judge. No author affiliation made an evaluated model.",
    "attribution": "Xueting Fang, Zehui Li, Yang Yang, Camilla Giovino, Shubh K. Patel, Shailly Prajapati, Vallijah Subasri, Caihua Shan (Zhejiang U., Imperial College London, Nanchang U., U. Toronto, UHN, Microsoft), 'KlinikeBench: Evaluating Language Models Beyond Diagnostic Accuracy', arXiv:2609.38480v2 (2026), CC BY 4.0. Changes: selected cells of Table 7 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 7 (results for 31 models on 333 tasks), 'Strict pass@1' column only (Diagnosis/Tools/Must-ask are components of the same gate). GPT-4o (no snapshot given; three catalogued snapshots) and DeepSeek-v4-pro/flash (preview vs final release not stated) not imported.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2609.38480v2",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2610.04118v1-longsocialbench",
    "name": "LongSocialBench: Do Long-Context LLMs Understand Online Discussion Threads? (arXiv:2610.04118v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2610.04118v1",
      "arxiv.org/abs/2610.04118v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2610.04118v1",
    "quote": "arXiv abs page for 2610.04118v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4: 'We evaluate 18 contemporary LLMs spanning closed-source frontier and strong open-weight systems ... Closed-source models include Claude Opus 4.7, Claude Sonnet 4.6, 4.5, and 4, GPT-5, and Gemini 2.5 Pro. Open-weight models include Mistral Large 3, Llama 4 Scout and Maverick, Llama 3.1 and 3.2, Qwen3-Next and Qwen3-32B, GLM-5, DeepSeek-R1, Gemma 3, and OpenAI's gpt-oss models.' Metrics: 'responses are deterministically parsed to an A-D choice and scored by exact match. Accuracy is computed over successfully parsed responses only ... Benchmark-level results report micro accuracy.' Appendix Table 11 lists the models (open weights from Hugging Face) and context caps. Authors (UIUC, Semantic Code Lab) made none of the models.",
    "attribution": "Xinyi Liu, Rinat Khaziev, Dilek Hakkani-Tür, Tarek F. Abdelzaher (University of Illinois Urbana-Champaign, Semantic Code Lab), 'LongSocialBench: Do Long-Context LLMs Understand Online Discussion Threads?', arXiv:2610.04118v1 (2026), CC BY 4.0. Changes: selected cells of Table 13 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 13 (per-model accuracy on three representative workflows), 'Full' column = full-context prompting micro accuracy on the 1,462 released items. CoT/LLM-as-retriever/Best/C-mac columns not used. Qwen3-Next-80B not imported (Instruct vs Thinking not stated).",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2610.04118v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2610.04432v2-video2world",
    "name": "Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied Videos (arXiv:2610.04432v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2610.04432v2",
      "arxiv.org/abs/2610.04432v2"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://arxiv.org/abs/2610.04432v2",
    "quote": "arXiv abs page for 2610.04432v2 links http://creativecommons.org/licenses/by-sa/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY-SA 4.0'. Sec. 4.1: 'We evaluate 9 coding-agent models on 222 instances, covering 189 video clips and 39 task families. Seven models run through OpenCode, while GPT-6 Astra and Fable-5.1 use Codex CLI and Claude Code, respectively. We evaluate each configuration as a complete agent system ... Each model makes one autonomous construction attempt per instance ... within a 180-minute wall-clock limit. The submission retained by the prescribed stopping rule is frozen for evaluation, without human correction or selection based on hidden scores.' Table A7 gives the exact model identifiers (OpenCode 1.18.29, openrouter/ prefix) and provider routes. Authors (Aether AI, UCSD) made none of the evaluated models.",
    "attribution": "Jinzhou Tang, Zijun Zhang, Jing Yang, Yuchen Yan, Kun Zhou, Lingjun Mao, Ruobing Han, Jinglin Cao, Wenpeng Xu, Lukun He, Minghao Fu, Fan Feng, Biwei Huang (Aether AI, UC San Diego), 'Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied Videos', arXiv:2610.04432v2 (2026), CC BY-SA 4.0. Changes: selected cells of Table 2 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 2 'V2WScore' column for the seven configurations that share the OpenCode 1.18.29 scaffold (Table A7). GPT-6 Astra (Codex CLI, 'Ultra') and Fable-5.1 (Claude Code, 'Max') use different scaffolds and are not imported, to keep one harness per run; the human-assisted reference row is not a model. Component columns not used.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2610.04432v2",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2610.12402v1-spacecast-bench",
    "name": "SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models (arXiv:2610.12402v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2610.12402v1",
      "arxiv.org/abs/2610.12402v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2610.12402v1",
    "quote": "arXiv abs page for 2610.12402v1 links http://creativecommons.org/licenses/by/4.0/ as the licence of this version (checked 2026-10-10); the HTML version header states 'License: CC BY 4.0'. Sec. 4.1: 'We evaluate 21 models in 3 groups: 9 proprietary models, 6 general-purpose open-weight models, and 6 spatially specialized models. The proprietary set includes Claude Sonnet 4.6 and Claude Sonnet 5; Gemini 3.5 Flash and Gemini 3 Flash; GPT-5.4 and GPT-5.6-Luna; Doubao 2.0 Pro; and Qwen3.6-Flash and Qwen3.7-Plus. The general-purpose open-weight set comprises InternVL3.5-8B, Qwen3-VL-4B, and Qwen3.5 models at 2B, 4B, 9B, and 27B ... All models are evaluated with a chain-of-thought prompt and deterministic decoding.' Metric: 'the overall score, defined as the unweighted macro-average of per-task accuracy across all 16 task types.' Authors (Zhejiang University) made none of the evaluated models.",
    "attribution": "Hongxing Li, Jinyue Su, Dingming Li, Wenqi Zhang, Weiming Lu, Jun Xiao, Yueting Zhuang, Yongliang Shen (Zhejiang University), 'SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models', arXiv:2610.12402v1 (2026), CC BY 4.0. Changes: selected cells of Table 1 transcribed into JSON; percentages divided by 100; model labels mapped to catalogue ids. No endorsement.",
    "scope": "Table 1 'Overall' column (macro-average accuracy over 16 task types) for catalogued proprietary and open-weight models. Not imported: Qwen3-VL-4B (Instruct vs Thinking not stated), Doubao 2.0 Pro and InternVL3.5-8B (not catalogued), spatial-specialist models; Human/Random rows are references.",
    "verdict": "allowed",
    "checked": "2026-10-10",
    "evidence_url": "https://arxiv.org/abs/2610.12402v1",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2610.00609v1-legal-research-bench",
    "name": "Legal Research Bench (Vals AI, arXiv:2610.00609v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/abs/2610.00609v1",
      "arxiv.org/html/2610.00609v1",
      "arxiv.org/pdf/2610.00609v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2610.00609v1",
    "quote": "arXiv abstract page for 2610.00609v1 (submitted 30 Sep 2026, only version) links 'http://creativecommons.org/licenses/by/4.0/' as the article licence. CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including commercially, subject to attribution.",
    "attribution": "Katrina Drozdov, Oliver Chen, Langston Nashold, Rayan Krishnan (Vals AI), Legal Research Bench: Measuring End-to-End Reliability in Long-Horizon Legal Research Agents, arXiv:2610.00609v1, CC BY 4.0. Table 3 and Table 4 all-pass cells; percentages converted to fractions; catalogue IDs mapped. No endorsement.",
    "scope": "Table 3 all-pass (agentic, with tools) and Table 4 all-pass (1-shot, no tools) for all 13 models; the authors ran every model themselves. Not imported: weighted pass (a partial-credit metric on the same answers), judge-calibration Table 2, per-category Tables 5-7. The live vals.ai leaderboard page is not covered by this licence.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2508.00828v1-finance-agent-benchmark",
    "name": "Finance Agent Benchmark (Vals AI, arXiv:2508.00828v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/abs/2508.00828v1",
      "arxiv.org/html/2508.00828v1",
      "arxiv.org/pdf/2508.00828v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2508.00828v1",
    "quote": "arXiv abstract page for 2508.00828v1 (submitted 20 May 2025, only version) links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; the HTML version header states 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including commercially, subject to attribution. Sec. 4.1: 'All models were evaluated under consistent conditions to ensure a fair comparison' (identical prompt, temperature 0 where supported, 16,384 max tokens).",
    "attribution": "Antoine Bigeard, Rayan Krishnan, Langston Nashold (Vals AI), Shirley Wu (Stanford), Finance Agent Benchmark: Benchmarking LLMs on Real-world Financial Research Tasks, arXiv:2508.00828v1, CC BY 4.0. Table 2 'Acc. (Class-Balanced)' cells; percentages converted to fractions; catalogue IDs mapped. No endorsement.",
    "scope": "Table 2 class-balanced accuracy (the paper's headline metric) on all 537 questions, highest reasoning row per model. Not imported: naive accuracy (same answers, different averaging), the non-thinking Claude 3.7 Sonnet row and the low-reasoning Grok 3 Mini Fast row (lower settings of a printed model), time/cost columns. Vals AI's later Finance Agent v1.1/v2 leaderboards on vals.ai are different versions and not covered by this licence.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2603.04601v3-vibe-code-bench",
    "name": "Vibe Code Bench (Vals AI, arXiv:2603.04601v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/abs/2603.04601v3",
      "arxiv.org/html/2603.04601v3",
      "arxiv.org/pdf/2603.04601v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2603.04601v3",
    "quote": "arXiv abstract page for 2603.04601v3 (v1 4 Mar 2026, v2 5 May 2026, v3 13 May 2026) links 'http://creativecommons.org/licenses/by/4.0/' as the licence of this version (v1 and v2 carry the same licence); the HTML version header states 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including commercially, subject to attribution. Sec. 4: 'We evaluate 16 models from 9 providers ... All models use vision if supported, the highest reasoning level available for that model, and the default temperature specified by the model provider.'",
    "attribution": "Hung Tran, Langston Nashold, Rayan Krishnan, Antoine Bigeard (Vals AI), Alex Gu (MIT), Vibe Code Bench: Evaluating AI Models on End-to-End Web Application Development, arXiv:2603.04601v3, CC BY 4.0. Table 4 accuracy cells; percentages converted to fractions; catalogue IDs mapped. No endorsement.",
    "scope": "Table 4 accuracy on the 50-app test split for 15 of 16 rows (the plain 'Claude Opus 4.6' row is a lower setting of the printed 'Claude Opus 4.6 Thinking'); model ids and reasoning settings from Table 15. Table 4 is identical in v1, v2 and v3. Not imported: per-difficulty/integration/domain breakdowns (Tables 5-7, 11-13: same answers), the six-app variance panel (Table 9), evaluator-agreement Table 10, ablation figures, cost/latency. The vals.ai Vibe Code Bench leaderboard and the vibe-code-bench-cais-2026-artifacts repo (no licence) are not covered.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2603.26996v1-formalproofbench",
    "name": "FormalProofBench (Vals AI, arXiv:2603.26996v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/abs/2603.26996v1",
      "arxiv.org/html/2603.26996v1",
      "arxiv.org/pdf/2603.26996v1"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2603.26996v1",
    "quote": "arXiv abstract page for 2603.26996v1 (submitted 27 Mar 2026, only version) links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; the HTML version header states 'License: CC BY 4.0'. CC BY 4.0 s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including commercially, subject to attribution. Sec. 3.2: 'We standardize our evaluation settings by using the default sampling settings from each model provider, and by using the native model provider rather than a third-party inference provider.' Sec. 5: 'Table 2 summarizes the overall results on models from OpenAI, Anthropic, xAI, Zhipu AI, Google, and DeepSeek.'",
    "attribution": "Nikil Ravi, Kexing Ying, Vasilii Nesterov, Rayan Krishnan, Elif Uskuplu, Bingyu Xia, Janitha Aswedige, Langston Nashold (Vals AI and co-authors), FormalProofBench: Can Models Write Graduate Level Math Proofs That Are Formally Verified?, arXiv:2603.26996v1, CC BY 4.0. Table 2 accuracy cells; percentages converted to fractions; catalogue IDs mapped. No endorsement.",
    "scope": "Table 2 accuracy (200 problems) for 11 catalogued models, highest reasoning row per model (DeepSeek V3.2 Thinking, not Nonthinking). Not imported: DeepSeek V3.2 (Nonthinking) (lower setting), GPT 5.1 Codex Max (not catalogued), cost/latency columns, tool-use figures. The vals.ai ProofBench leaderboard is not covered by this licence.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "card3-liquidai-lfm2-5-2-6b",
    "name": "LiquidAI/LFM2.5-2.6B model card",
    "kind": "first-party",
    "match": [
      "huggingface.co/LiquidAI/LFM2.5-2.6B/blob/654f9463ce32b05d0429d76fe1f580b27d4c1ac0/README.md#benchmarks"
    ],
    "exact": true,
    "license": "LFM Open License v1.0",
    "license_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/blob/654f9463ce32b05d0429d76fe1f580b27d4c1ac0/LICENSE",
    "terms_url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/blob/654f9463ce32b05d0429d76fe1f580b27d4c1ac0/LICENSE",
    "quote": "Same licence and treatment as the ledger's card-liquidai-lfm2-5-1-2b-thinking and card3-liquidai-lfm2-1-2b: the card front matter (license: other, license_name: lfm1.0, license_link: LICENSE) links the LFM Open License v1.0 ('TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION'), which covers documentation and grants reproduction, adaptation, display and distribution subject to its terms (Section 5 commercial revenue / Legal Entity conditions apply to recipients). The match carries the '#benchmarks' fragment because the same README URL is already matched by first-party-liquid-ai (facts only: list price/context); the fragment keeps these score rows on this licence entry.",
    "conditions": "LFM Open License v1.0 includes commercial revenue and Legal Entity conditions. Downstream users must assess their own eligibility; this compilation does not remove these restrictions.",
    "attribution": "Liquid AI. LiquidAI/LFM2.5-2.6B model card. Source: https://huggingface.co/LiquidAI/LFM2.5-2.6B/blob/654f9463ce32b05d0429d76fe1f580b27d4c1ac0/README.md. Licensed under LFM Open License v1.0. Modified for whatmodel.sh: selected cells, percentages to fractions. No endorsement.",
    "scope": "'### Benchmarks' table: LFM2.5-2.6B own column and the gemma-4-E2B-it, gemma-4-E4B-it, Qwen3.5-4B, Qwen3.5-9B comparator columns (two-decimal values that differ from the makers' published figures, read as Liquid's own runs) for AIME25, LiveCodeBenchv6, IFBench, Multi-IF, BFCLv4, Claw-Eval average (EN), τ³-Bench Banking. AA-Omni-Public rows excluded.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "huggingface.co"
  },
  {
    "id": "nsanku-tts-benchmark-mit",
    "name": "nsanku TTS Benchmark (GhanaNLP community Space)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/ghananlpcommunity/nsanku-tts-benchmark/blob/41b828ed6803bd052b1bef94bb6802fecd2edf8c/bundled_data.json"
    ],
    "exact": true,
    "license": "MIT",
    "license_url": "https://huggingface.co/spaces/ghananlpcommunity/nsanku-tts-benchmark/blob/41b828ed6803bd052b1bef94bb6802fecd2edf8c/README.md",
    "terms_url": "https://huggingface.co/spaces/ghananlpcommunity/nsanku-tts-benchmark/blob/41b828ed6803bd052b1bef94bb6802fecd2edf8c/README.md",
    "quote": "Space card YAML at the pinned revision: 'license: mit'. README: 'Multi-metric Text-to-Speech benchmark for Ghanaian languages ... Per-language ASR judges transcribe each clip; scored against the sentence read.' The Space holds only README.md, bundled_data.json, index.html and style.css; no contrary terms. The linked GitHub repo (GhanaNLP/nsanku-tts-benchmark) has no licence and is not used.",
    "attribution": "GhanaNLP community, nsanku TTS Benchmark (huggingface.co/spaces/ghananlpcommunity/nsanku-tts-benchmark), MIT. Selected per-language CER values averaged over 7 languages; model IDs mapped; no endorsement.",
    "scope": "bundled_data.json per-language 'cer' for the entries Google/gemini-3.1-flash-tts-preview, k2-fsa/OmniVoice-ref and ghananlpcommunity/ghana-tts-72k-ref.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "huggingface.co"
  },
  {
    "id": "speech-sglang-omni-cookbooks-apache",
    "name": "SGLang-Omni serving cookbooks (SGLang-Omni team measurements)",
    "kind": "leaderboard",
    "match": [
      "github.com/sgl-project/sglang-omni/blob/9641219b320597393d915afb8d03352c629d94a5/docs/cookbook/moss_tts.md",
      "github.com/sgl-project/sglang-omni/blob/9641219b320597393d915afb8d03352c629d94a5/docs/cookbook/qwen3_tts.md"
    ],
    "exact": true,
    "license": "Apache-2.0",
    "license_url": "https://github.com/sgl-project/sglang-omni/blob/9641219b320597393d915afb8d03352c629d94a5/LICENSE",
    "terms_url": "https://github.com/sgl-project/sglang-omni/blob/9641219b320597393d915afb8d03352c629d94a5/LICENSE",
    "quote": "Repository root LICENSE at the pinned commit is the Apache License 2.0 (GitHub licence detection: Apache-2.0) and covers the documentation files. The cookbook benchmark sections are the SGLang-Omni team's own serving measurements ('reference numbers tabulated in benchmarks/eval/benchmark_tts_seedtts.py'; 'Hardware: 1× H200 SXM'). Same repository as the existing speech-higgs3-sglang-apache entry (Boson AI tables at 0aafd76), which is a different contributor and commit.",
    "attribution": "SGLang-Omni contributors, sgl-project/sglang-omni docs/cookbook (moss_tts.md, qwen3_tts.md) at 9641219b320597393d915afb8d03352c629d94a5. Apache License 2.0. Changes: selected corpus WER/CER cells for catalogued models, percentages divided by 100. No endorsement.",
    "scope": "Seed-TTS-Eval corpus WER/CER from the 'Benchmark Results' sections whose protocol matches an existing benchmark id: moss_tts.md (MOSS-TTS-v1.5, cloning, Whisper-large-v3 EN / paraformer-zh ZH) and qwen3_tts.md '0.6B Base' (cloning, Whisper-large-v3 EN). Preset-voice, Qwen3-ASR-scored and x-vector experiment tables are not used.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "github.com"
  },
  {
    "id": "paper-2409.00696v3-polyrating",
    "name": "Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2409.00696v3"
    ],
    "exact": true,
    "license": "CC-BY-4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2409.00696v3",
    "quote": "arXiv abstract page for 2409.00696v3 links 'http://creativecommons.org/licenses/by/4.0/' as the article licence; CC BY s.2(a)(1) grants a worldwide, royalty-free licence to reproduce and Share the Licensed Material, including for commercial purposes, subject to attribution.",
    "attribution": "Authors of arXiv:2409.00696 (Polyrating), CC BY 4.0. No endorsement.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "evidence_url": "https://arxiv.org/abs/2409.00696v3",
    "icon": "arxiv.org",
    "scope": "Table 7 Rating column (not the modifier columns, not unidimensional Table 8) for 96 rows: 82 matched to catalogue models by exact id/name, 13 whose Arena key is mapped exactly as our lmarena-text rows map the same key, and reka-core-20240501. Unmapped: preview/product keys (gemini-advanced-0514, bard-jan-24-gemini-pro, gemini-1.5-pro-api-0514/-0409-preview, gemini-1.5-flash-api-0514, gemini-pro, reka-flash-preview-20240611, reka-flash-21b-20240226-online, mistral-next, pplx-*-online) and uncatalogued models (yi-large(-preview), claude-1, claude-instant-1, qwen-max-0428, glm-4-0116, gpt-3.5-turbo-0314)."
  },
  {
    "id": "report-2602.04705v1",
    "name": "Baidu - ERNIE 5.0 Technical Report (arXiv:2602.04705v1)",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2602.04705v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2602.04705v1",
    "quote": "arXiv abstract page for 2602.04705 (only version v1, submitted 2026-02-04) links http://creativecommons.org/licenses/by/4.0/ (checked 2026-10-11); the HTML header reads 'License: CC BY 4.0'. Sec. 6: 'We conduct systematic evaluations of ERNIE 5.0 against state-of-the-art models across a wide range of text (Sec. 6.1), vision (Sec. 6.2), and audio benchmarks (Sec. 6.3) with internal evaluation framework, ERNIE-Eval'.",
    "attribution": "Baidu ERNIE Team. ERNIE 5.0 Technical Report. arXiv:2602.04705v1. CC BY 4.0. Selected numerical cells transcribed; percentages divided by 100; 'ChartXiv' read as CharXiv. No endorsement.",
    "scope": "Table 3 rows ChartXiv-RQ and ChartXiv-DQ: ERNIE 5.0 column (lab) and the ChartXiv-DQ cells of Gemini 2.5-Pro, GPT-5 (High) and Gemini 3-Pro (independent; covered by the Sec. 6 ERNIE-Eval statement and not published by their makers). Not used: competitor ChartXiv-RQ cells (GPT-5 81.10 and Gemini 3-Pro 81.40 equal the makers' own published figures, so they look copied), the unsized Qwen3-VL Thinking column, other rows.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-opengvlab-scores",
    "name": "OpenGVLab (Shanghai AI Laboratory) vendor-reported scores for its own InternVL models (InternVL2.5 report, arXiv 2412.05271v5)",
    "kind": "first-party",
    "match": [
      "arxiv.org/html/2412.05271v5"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "license_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "terms_url": "https://arxiv.org/abs/2412.05271v5",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. The arXiv abstract page for 2412.05271v5 links http://arxiv.org/licenses/nonexclusive-distrib/1.0/ (not an open licence), so only rule (b) applies.",
    "attribution": "OpenGVLab / Shanghai AI Laboratory (Zhe Chen et al.), 'Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling', arXiv:2412.05271v5; own-model figures hand-recorded 2026-10-11; percentages converted to fractions. Lab-reported. No endorsement.",
    "scope": "Table 7 row 'InternVL2-4B' CharXiv (RQ / DQ) only. InternVL2-26B row excluded (its 33.4 / 62.4 equals the CharXiv authors' README figure, already recorded from that source); InternVL2.5 rows not catalogued; other labs' rows excluded.",
    "verdict": "allowed",
    "checked": "2026-10-11"
  },
  {
    "id": "paper-2406.18521v1-charxiv",
    "name": "CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs (arXiv:2406.18521v1)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2406.18521v1",
      "arxiv.org/abs/2406.18521v1"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://arxiv.org/abs/2406.18521v1",
    "quote": "arXiv abstract page for 2406.18521 (only version v1) links http://creativecommons.org/licenses/by-sa/4.0/ as the article licence (checked 2026-10-11). Owner note 2026-10-09: 'The CharXiv arXiv paper is CC BY-SA 4.0 and usable.' Table 11 'Run configurations for all models' documents the authors' own runs of every model.",
    "attribution": "Zirui Wang et al. (Princeton NLP), CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs, arXiv:2406.18521v1, CC BY-SA 4.0. Test-set overall accuracies from Appendix C Table 4; percentages converted to fractions; model names mapped via Table 11. Adapted rows are distributed under CC BY-SA 4.0 (data/extracts). No endorsement.",
    "scope": "Appendix C Table 4 test-split RQ All and DQ All for the 7 models whose Table 11 checkpoint is catalogued (gpt-4o-2024-05-13, gpt-4-turbo-2024-04-09, Claude 3 Sonnet/Haiku/Opus, reka-flash-20240226, llava-v1.6-34b). Not used: Reka Core (reka-core-20240415 not catalogued), Gemini 1.0 Pro Vision, Qwen VL Max/Plus (no snapshot), Reka Edge and open models not in the catalogue.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "first-party-minimax-scores",
    "name": "MiniMax vendor-reported benchmark scores for its own models (Hugging Face model cards)",
    "kind": "first-party",
    "match": [
      "huggingface.co/MiniMaxAI/MiniMax-M2.1/blob/cd97f59135f37b2a6bf09356e485d5e4aeb7dc9c/README.md"
    ],
    "exact": true,
    "license": "Factual vendor-reported score (not licensed content)",
    "terms_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1/blob/cd97f59135f37b2a6bf09356e485d5e4aeb7dc9c/README.md",
    "quote": "Owner rule (b), approved 2026-10-07: a vendor's own published score for its own model may be recorded by hand one figure at a time with the exact URL, date and printed settings; other labs' cells excluded; rows labelled by: 'lab'. (The card's repo licence is 'modified-mit' (MIT plus a condition to display 'MiniMax M2.1' in commercial products); it is not relied on here.)",
    "attribution": "MiniMax, MiniMax-M2.1 model card (@cd97f59), own-model figures hand-recorded 2026-10-11; percentages converted to fractions. Lab-reported. No endorsement.",
    "scope": "Last benchmark table, row '𝜏²-Bench Telecom', columns MiniMax-M2.1 and MiniMax-M2 (MiniMax's own AA-methodology runs). Other labs' columns excluded.",
    "verdict": "allowed",
    "checked": "2026-10-11"
  },
  {
    "id": "paper-2607.00248v1-seed2-0",
    "name": "ByteDance Seed - Seed2.0 Model Card (arXiv:2607.00248v1)",
    "kind": "first-party",
    "match": [
      "arxiv.org/pdf/2607.00248v1",
      "arxiv.org/abs/2607.00248v1"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2607.00248v1",
    "quote": "arXiv abstract page for 2607.00248 (only version v1, submitted 2026-06-30) links http://creativecommons.org/licenses/by/4.0/ (checked 2026-10-11). No HTML rendering; values read from the PDF text layer.",
    "attribution": "ByteDance Seed, Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity, arXiv:2607.00248v1, CC BY 4.0. Own-model cells transcribed; percentages converted to fractions. No endorsement.",
    "scope": "Own-model cells only (Seed2.0 Pro / Lite), Sec. 3.3 agentic tables, row τ2-Bench (telecom). Competitor cells excluded: Sec. 3.3 takes 'the maximum of the score reported in the official documentation and the score obtained in our tests', so they mix vendor figures and ByteDance runs.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "paper-2509.14128v2-canary-parakeet",
    "name": "Canary-1B-v2 & Parakeet-TDT-0.6B-v3: Efficient and High-Performance Models for Multilingual ASR and AST (NVIDIA, arXiv:2509.14128v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2509.14128v2",
      "arxiv.org/abs/2509.14128v2"
    ],
    "exact": true,
    "license": "CC BY-SA 4.0",
    "license_url": "https://creativecommons.org/licenses/by-sa/4.0/",
    "terms_url": "https://arxiv.org/abs/2509.14128v2",
    "quote": "arXiv abstract page for 2509.14128v2 links http://creativecommons.org/licenses/by-sa/4.0/ (checked 2026-10-11; v1 has the same licence); the HTML header states 'License: CC BY-SA 4.0'. Sec. 6.1: 'For comparative analysis, we consider several strong baselines ...' and normalized WER with the Open ASR normalizers including 'the newly added multilingual normalizer for non-English test sets'.",
    "attribution": "Monica Sekoyan et al. (NVIDIA), Canary-1B-v2 & Parakeet-TDT-0.6B-v3, arXiv:2509.14128v2, CC BY-SA 4.0. Per-language WERs from Appendix Tables 11-13 averaged (unweighted) into the benchmark's language sets; percentages converted to fractions. Adapted rows are distributed under CC BY-SA 4.0 (data/extracts). No endorsement.",
    "scope": "Appendix Tables 11-13 (FLEURS, MLS, CoVoST2 per-language WER) for models with every language of the benchmark set: whisper-large-v3 (all three), seamless-m4t-v2-large (MLS only; FLEURS and CoVoST2 lack Latvian), Parakeet-TDT-0.6B-v3 and Canary-1B-v2 (own, all three). Not used: Table 5 (Open ASR Leaderboard figures), Voxtral-Mini and Phi-4-multimodal (partial language coverage), seamless-m4t-medium (not catalogued), AST/COMET tables.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "speech-stt-qcri-aura-ccby",
    "name": "QCRI Arabic + English ASR leaderboard (Hugging Face Space QCRI/AURA-Arabic-English-ASR-leaderboard)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/spaces/QCRI/AURA-Arabic-English-ASR-leaderboard/blob/208ff13439f377131207a2b32bf73364ba0e2d0c/leaderboard.tsv"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://huggingface.co/spaces/QCRI/AURA-Arabic-English-ASR-leaderboard/blob/208ff13439f377131207a2b32bf73364ba0e2d0c/README.md",
    "quote": "Space README front matter at @208ff13: 'license: cc-by-4.0'; the committed leaderboard.tsv is the data the Space renders. index.html: 'All reported WER/CER scores use the same bilingual Arabic–English text normalizer before scoring.'",
    "attribution": "QCRI (Qatar Computing Research Institute), Bilingual Arabic-English ASR Leaderboard, leaderboard.tsv @208ff13, CC BY 4.0. Average WER converted to a fraction; model labels mapped to catalogue ids. No endorsement.",
    "scope": "'Average WER' column for rows whose label maps to one catalogued speech model (cohere-transcribe-arabic-07-2026, qwen3_asr_1_7b, qwen3_asr_0_6b). Not used: QCRI's own AURA rows (not catalogued), 'whisper3' (size not stated), the Nemotron streaming variants (checkpoint not stated), Gemma-4 / Qwen2.5-Omni (LLM-category models), per-set columns.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "huggingface.co"
  },
  {
    "id": "paper-2508.21376v2-ahelm",
    "name": "AHELM: A Holistic Evaluation of Audio-Language Models (Stanford CRFM, arXiv:2508.21376v2)",
    "kind": "paper",
    "match": [
      "arxiv.org/html/2508.21376v2",
      "arxiv.org/abs/2508.21376v2"
    ],
    "exact": true,
    "license": "CC BY 4.0",
    "license_url": "https://creativecommons.org/licenses/by/4.0/",
    "terms_url": "https://arxiv.org/abs/2508.21376v2",
    "quote": "arXiv abstract page for 2508.21376v2 links http://creativecommons.org/licenses/by/4.0/ (checked 2026-10-11). The paper is a benchmark (HELM framework) in which the authors run every model themselves: 'To guarantee equitable and reliable comparisons among ALMs, we standardize the inference parameters by setting the model temperature to 0 and the maximum number of output tokens to 200'; prompts and raw outputs are released.",
    "attribution": "Tony Lee, Haoqin Tu, Chi Heem Wong et al. (Stanford CRFM), AHELM: A Holistic Evaluation of Audio-Language Models, arXiv:2508.21376v2, CC BY 4.0. Selected per-scenario scores from Appendix Tables A8-A19; model identifiers mapped via Table 3. No endorsement.",
    "scope": "Appendix Tables A8 (VoxCeleb2, VocalSound, LibriSpeech), A9 (AIR-Bench Foundation), A10 (CoRe-Bench), A11 (MELD, MUStARD), A15 (FLEURS, MLS WER) and A19 (Robust Speech Bench) for the 14 audio-language models (not the ASR+GPT-4o baseline pipelines). Not used: GPT-4o-judge critique scores, mean win rates, CoVoST-2 BLEU, bias/fairness/toxicity/safety tables.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "arxiv.org"
  },
  {
    "id": "hallucinations-leaderboard-results-apache",
    "name": "Hallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/datasets/hallucinations-leaderboard/results/blob/30f759225f89edcc18cb403b93f27bf4231223f7/"
    ],
    "exact": false,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/datasets/hallucinations-leaderboard/results/blob/30f759225f89edcc18cb403b93f27bf4231223f7/README.md",
    "quote": "Dataset card front matter at @30f7592: 'license: apache-2.0'. Each file is an EleutherAI lm-evaluation-harness results JSON written by the leaderboard backend (config.model, model_args 'pretrained=<repo>,revision=main,...', task versions), i.e. the leaderboard's own runs.",
    "attribution": "Hallucinations Leaderboard (Hong, Minervini et al.), hallucinations-leaderboard/results @30f7592, Apache-2.0. Selected task scores (nq8, tqa8, truthfulqa_mc2, halueval_qa), latest run per model and task; model repos mapped to catalogue ids. No endorsement.",
    "scope": "Files under the pinned revision for the 36 catalogued model repos; tasks nq8 (NQ-Open 8-shot EM), tqa8 (TriviaQA 8-shot EM), truthfulqa_mc2 and halueval_qa; latest run per model and task. Other tasks (summarisation ROUGE/factKB, FEVER, RACE, SQuADv2, MemoTrap, IFEval, SelfCheckGPT, ...) not used.",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "huggingface.co"
  },
  {
    "id": "openlifescienceai-results-apache",
    "name": "Open Medical-LLM Leaderboard results (HF dataset openlifescienceai/results)",
    "kind": "leaderboard",
    "match": [
      "huggingface.co/datasets/openlifescienceai/results/blob/62d7322cf11bf622399f7f54e5a7620ec6ef013d/"
    ],
    "exact": false,
    "license": "Apache-2.0",
    "license_url": "https://www.apache.org/licenses/LICENSE-2.0",
    "terms_url": "https://huggingface.co/datasets/openlifescienceai/results",
    "quote": "Dataset licence apache-2.0 (Hugging Face card metadata at @62d7322). Only lm-evaluation-harness outputs written by the leaderboard backend (config.model_args 'pretrained=<repo>,revision=main,dtype=...') are used; hand-entered files (Flan-PaLM, GPT-4, Med-PaLM 2, Gemini, GPT-3.5, the 2024-03-14_04-32-21 seed files and template-timestamp files) are excluded as possibly copied.",
    "attribution": "Open Medical-LLM Leaderboard (Open Life Science AI), openlifescienceai/results @62d7322, Apache-2.0. MedQA (4 options), MedMCQA and PubMedQA accuracies from harness result files; model repos mapped to catalogue ids. No endorsement.",
    "scope": "Harness-generated result files for 18 catalogued model repos (latest file per repo); tasks medqa_4options, medmcqa, pubmedqa. The six MMLU medical subsets are not used (parts of MMLU).",
    "verdict": "allowed",
    "checked": "2026-10-11",
    "icon": "huggingface.co"
  }
]