whatmodel.sh / LongCat 2.5 (1.6T-A48B)

LongCat 2.5 (1.6T-A48B) is an open-weights language model from Meituan, released November 2026. Its published size is 1600B parameters (48B active per token). Licence: mit.

Catalogue identity notes

Open BF16 weights on HF (meituan-longcat/LongCat-2.5, repo created 2026-11-04, 3,562 GB) plus Meituan's W4A4 export (meituan-longcat/LongCat-2.5-W4A4, 1,336 GB, INT4 experts for domestic accelerators; same card). API: LongCat API change log 'Version: 2026-11-02 - LongCat-2.5 Now Available'; distinct from LongCat-2.5-Preview (API only, 2026-09-25). Native multimodal (image input), hybrid thinking/non-thinking. The card's only deployment route is NPU via SGLang-FluentLLM (longcat-2.5-npu branch); no GPU or consumer command and no official Ollama/GGUF entry (ollama.com/library/longcat-2.5 404), so local is null. API price: Pricing page observed 2026-11-12 shows only a limited-time discounted price (US$0.30 uncached input / US$1.20 output per 1M tokens) and no standard list price, so no price recorded.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

Card: '1.6 trillion total parameters and approximately 48 billion activated per token' (nominal). HF safetensors metadata at 60032d9 totals 1,779,752,648,320 parameters (BF16 + F32) across 3,562.0 GB of shards including the multimodal parts; the card does not reconcile the two.

No comparable score yet. Its published results are below, but they don't overlap enough with other models to calculate a score.

$ how it compares

Scored LLMs released closest to LongCat 2.5 (1.6T-A48B), out of 1206 with a score, then more from Meituan.

#modelscorewhereRAMreleased
1Claude Fable 5.2 Anthropic89±1api $16.00/M2026-11
14Gemini 4 Hydrogen Google79±2api $3.00/M2026-11
22Hy4 Tencent78±1api price unknown2026-11
32Muse Ember Meta76±1api $0.20/M2026-11
37Inkling 2 Small (594B-A17B) Thinking Machines75±1api price unknown2026-11
more from Meituan
83LongCat 2.0 (1.6T-A48B) Meituan71±2api $1.30/M +local~2155GB2026-06
125LongCat-Flash-Chat-2602-Exp Meituan68±3comparison only
140LongCat-Flash-Thinking (560B-A27B) Meituan67±1comparison only2025-09

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasksnot ranked on this task yet · 7 tests
testresultvs the others in its rundifficultycountswhy
Terminal-Bench 2.1 lab88.4% ✎▁·▁·▁▁▄▃▅█▆█ beats 63/68—not counted yet: no other model in the same run to compare it with
Humanity's Last Exam with tools lab63.1% ✎▁·▁▂▁▂▃▂█▅▄▁ beats 55/64—not counted yet: no other model in the same run to compare it with
Terminal-Bench 4.0 lab38.5% ✎▂·▁··▂▄▄▅█▇▄ beats 24/41—not counted yet: no other model in the same run to compare it with
MCP-Atlas (public set) lab83.1% ✎▂·▁▂▂·▁▃█▆▆▃ beats 40/41—not counted yet: no other model in the same run to compare it with
DeepSWE v1.1 (resolved) lab64.6% ✎▂····▃▁▁▄█▆█ beats 14/36—not counted yet: no other model in the same run to compare it with
Agents' Last Exam (metric/harness unspecified) lab38.5% ✎▂·▂·█▄▂▃▂▂▄▄ beats 9/15—not counted yet: no other model in the same run to compare it with
JobBench lab65.4% ✎▄·▄▄···█···▄ beats 2/5—not counted yet: no other model in the same run to compare it with
codingnot ranked on this task yet · 1 test
testresultvs the others in its rundifficultycountswhy
DeepSWE v1.1 (resolved) lab64.6% ✎▂····▃▁▁▄█▆█ beats 14/36—not counted yet: no other model in the same run to compare it with
web searchnot ranked on this task yet · 1 test
testresultvs the others in its rundifficultycountswhy
Humanity's Last Exam with tools lab63.1% ✎▁·▁▂▁▂▃▂█▅▄▁ beats 55/64—not counted yet: no other model in the same run to compare it with
other tests5 not in a task · 0.0% of its score
testresultvs the others in its rundifficultycountswhy
NL2Repo lab60.5% ✎▂▂▂·▃▃▅█▃▃▆▃ beats 21/24—not counted yet: no other model in the same run to compare it with
AutomationBench v1.0.6 lab51.8% ✎▂···▂·▂▂··█▆ beats 5/10—not counted yet: no other model in the same run to compare it with
Claw-Eval Pass^3 lab78.3% ✎██▄▄▄·▄▄···▄ beats 9/9—not counted yet: no other model in the same run to compare it with
ResearchRubrics lab81.0% ✎█▃▅·▃······▃ beats 7/7—not counted yet: no other model in the same run to compare it with
ProgramBench (Almost@1) lab25.5% ✎██·█·█···█·█ beats 2/5—not counted yet: no other model in the same run to compare it with

$ where to run it

Through an API

api · LongCat API Platform (Meituan) · price unknown · longcat.chat ↗

Model repository and documentation: huggingface.co/meituan-longcat ↗

$ published results · 12 results, not enough overlap for a score
testresultsource and conditions
Terminal-Bench 2.188.4%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'Terminal Bench v2.1'. Footnote 1: Claude Code harness (v2.1.280), max_seq_length = 512K; 'Terminal Bench v2.1 results are averaged over three independent runs.' Printed 88.4.

Terminal-Bench 4.038.5%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'Terminal Bench v4.0'. Footnote 1: Claude Code harness (v2.1.280), max_seq_length = 512K. Printed 38.5.

ProgramBench (Almost@1)25.5%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'Program Bench (Almost Resolved)'. Row label 'Program Bench (Almost Resolved)'; footnote 1: Claude Code harness (v2.1.280). The row's DeepSeek-V4.1-Flash cell (20.3*) equals the DB's DeepSeek programbench-almost1 row. Printed 25.5.

DeepSWE v1.1 (resolved)64.6%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'DeepSWE (v1.1)'. Footnote 2: mini-swe-agent, temperature 1.0, top_p 0.95, max_seq_length 1M, averaged over three runs. Printed 64.6.

NL2Repo60.5%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'NL2Repo'. Footnote 1: Claude Code harness (v2.1.280), max_seq_length = 512K. Printed 60.5.

AutomationBench v1.0.651.8%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'AutomationBench (v1.0.6)'. Footnote 7: 'AutomationBench (public v1.0.6): All 600 public tasks, evaluated with the benchmark's official evaluation code; temperature = 1.0, top_p = 0.95.' Printed 51.8.

Humanity's Last Exam with tools63.1%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'HLE (w/ tools)'. Footnote 5: 'Full set, evaluated with Claude Code (v2.1.116); maximum generation length 256K tokens.' Printed 63.1.

Agents' Last Exam (metric/harness unspecified)38.5%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'ALE (v1.1)'. Row 'ALE (v1.1)'; footnotes give no harness or metric for ALE. Printed 38.5.

ResearchRubrics81.0%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'ResearchRubrics'. No footnote for this row. Printed 81.0.

MCP-Atlas (public set)83.1%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'MCP Atlas'. Footnote 6: 'Reasoning effort set to max, temperature = 0.7, top_p = 0.95, 256K context.' Split (public / v1 public500) not stated. Printed 83.1.

JobBench65.4%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'JobBench'. Footnote 8: 'JobBench (Main 65): Claude Code harness (v2.1.280); scoring follows the official JobBench LLM-as-judge protocol ... with gemini-3.6-flash as the judge'. Printed 65.4.

Claw-Eval Pass^378.3%meituan-longcat/LongCat-2.5 model card · maker-reported

Meituan LongCat own evaluation in the pinned meituan-longcat/LongCat-2.5 model card (sha 60032d9), Evaluation Results table, column 'LongCat-2.5'. Card: 'Scores marked with * are cited from public technical reports, blog posts, or official leaderboards; unmarked scores are measured in-house and are indicative only.' LongCat-2.5 cells are unmarked. Footnotes: 'Unless otherwise specified, all LongCat-2.5 results use enable_thinking=on, temperature=1.0, and top_p=0.95.' Other models' columns not used. Row 'Claw-Eval (Pass^3)'. Row 'Claw-Eval (Pass^3)'; no footnote. Printed 78.3.

$ improve this page

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json