whatmodel.sh / Inkling (975B-A41B)
Inkling (975B-A41B)
↗ share
+ contribute
Inkling (975B-A41B) is an open-weights language model from Thinking Machines, released July 2026. Its published size is 975B parameters (41B active per token) and a 1M-token context window. Licence: apache-2.0.
$ how it compares
The LLMs ranked either side of Inkling (975B-A41B), out of 828 with a score, then more from Thinking Machines.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 94 |  Qwen3-Max (2026-01-23) | 70±1 | comparison only | ||
| 95 |  MiMo-V2.5-Pro (1T-A42B) | 70±1 | api $0.65/M +local | ||
| 96 | Inkling (975B-A41B) | 70±1 | api $1.73/M +local | ||
| 97 |  Muse Glimmer 30B | 70±1 | api $0.64/M +local | ||
| 98 |  Nemotron 3 Ultra 550B-A55B | 69±1 | api $1.35/M +local | ||
| more from Thinking Machines | |||||
| 90 | Inkling Small (266B-A12B) | 70±1 | api $0.64/M +local | ||
$ what it's good at
agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #79 of 300 · 23% of its score · 10 tests
Beats GPT-5, Grok 4.20 (0309 reasoning), GPT-5.1 here, though they rank higher overall. Trails MiniMax-M2.7, MiMo-V2.5 (310B-A15B), Qwen3.6 27B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena Agent (October 2026 aggregate IPS) | -0.1086 IPS ✎ | ▁▁▃▄▂▃▇█▂▂▁▃ beats 2/49 | 9.1% |
| SWE-bench Verified lab | 77.6% ✎ | ▁▁▁▁▁▁▂▄█▆▁▁ beats 114/131 | 2.8% |
| τ³-bench banking_knowledge (v1.0.1 grading) | 25.0% ✎ | ▃▃▂▃·▆▃▂█▅▅▂ beats 8/25 | 2.7% |
| Terminal-Bench 2.1 lab | 63.8% ✎ | ▁·▁·▁▁▄▃▄█▃▇ beats 26/56 | 2.6% |
| MCP-Atlas (public set) lab | 78.8% ✎ | ▂·▁▂▁·▁▃█▆▆▂ beats 34/37 | 1.6% |
| Humanity's Last Exam with tools lab | 46.0% ✎ | ▂▂▂▁▄▂▃▆█▇▂▆ beats 21/54 | 1.3% |
| BrowseComp with context management lab | 77.1% ✎ | ▅·▅▃·█▃·█▅·▃ beats 6/14 | 1.2% |
| Toolathlon-Verified lab | 45.5% ✎ | ▂·▂▂▅▃···▂██ beats 3/18 | 0.8% |
2 more tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| SWE-bench Pro lab | 54.3% ✎ | ▅▃▃·▃·▅▅▅█▃▃ beats 5/15 | 0.7% |
| τ³-bench Banking lab | 23.7% ✎ | ▂▂·▅█▂▂▃▂▅▂▃ beats 17/20 | 0.6% |
coding▁▁▁▃▆██▇▆▄▃▁ #80 of 458 · 13% of its score · 4 tests
Beats Grok 4.20 (0309 reasoning), GPT-5.4 mini, GPT-5.1 here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, GLM-4.7, Gemini 3.5 Flash-Lite here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena WebDev (Code Arena) | 1412.6 ✎ | ▁▂▄▃▅█▄▇▆▄▁▁ beats 48/119 | 8.1% |
| SWE-bench Verified lab | 77.6% ✎ | ▁▁▁▁▁▁▂▄█▆▁▁ beats 114/131 | 2.8% |
| SciCode (level unspecified) lab | 46.1% ✎ | ▁····▃▄▄▄█▆▂ beats 31/34 | 1.2% |
| SWE-bench Pro lab | 54.3% ✎ | ▅▃▃·▃·▅▅▅█▃▃ beats 5/15 | 0.7% |
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #63 of 393 * · 4.5% of its score · 1 test
Beats GPT-5.2, Qwen3.8 27B, LongCat 2.0 (1.6T-A48B) here, though they rank higher overall. Trails Gemma 4 31B, Gemini 2.5 Pro, Claude Sonnet 4.5 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena (text) | 1441.4 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 299/379 | 4.5% |
math▁▂▃▅▆▅▄█▄▁▁▁ #89 of 398 · 37% of its score · 5 tests
Beats Claude Sonnet 4.6, Claude Opus 4.5, MiniMax-M3 here, though they rank higher overall. Trails Gemini 2.5 Flash, Granite 4.2 30B, INTELLECT-3 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| FrontierMath tiers 1–3 v2 | 33.3% ✎ | ▂··▁▁▃▄▄█▅▃▃ beats 21/82 | 21% |
| FrontierMath tier 4 v2 | 4.9% ✎ | ▁▂▃▂█▅▂▂▁▁▁▁ beats 10/63 | 7.7% |
| HMMT February 2026 lab | 86.3% ✎ | ▃▃▂▃·██▆█▅▃▅ beats 26/33 | 3.0% |
| AIME 2026 lab | 97.1% ✎ | ▁▁▁▂▁▂▆█▆▃·▁ beats 48/49 | 2.9% |
| OTIS mock AIME 2024–2025 | 88.9% ✎ | ▃▃▄▃▃▃▇▇█▃▂▅ beats 140/193 | 2.2% |
vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #65 of 147 * · 3.0% of its score · 2 tests
Beats DeepSeek V4 Flash Vision (Exp), GPT-5, Inkling Small (266B-A12B) here, though they rank higher overall. Trails Gemma 4 26B-A4B, Claude Haiku 4.5, Qwen3.6 Flash here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CharXiv (reasoning questions) lab | 78.1% ✎ | ▂▅▃▅▅▆▆▅▆█▆▂ beats 72/100 | 1.6% |
| MMMU Pro lab | 73.5% ✎ | ▂▃▁▃▆▇▃▇▇██▅ beats 35/52 | 1.4% |
web search█··········█ #2 of 2 * · 2.5% of its score · 2 tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Humanity's Last Exam with tools lab | 46.0% ✎ | ▂▂▂▁▄▂▃▆█▇▂▆ beats 21/54 | 1.3% |
| BrowseComp with context management lab | 77.1% ✎ | ▅·▅▃·█▃·█▅·▃ beats 6/14 | 1.2% |
other tests9 not in a task · 23% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GPQA Diamond +1 | 88.3% ✎ | ▁▁▁▁▂▄▆▄▅▇█▅ beats 162/208 | 9.4% |
| SimpleQA Verified +1 | 40.3% ✎ | ▁▁▃▂▃▃▆▆█▃▂▅ beats 33/78 | 2.7% |
| Chess Puzzles (Epoch) | 21.0% ✎ | █▁▁▁▁▂▁▄▂▂▁▁ beats 85/135 | 2.4% |
| Humanity's Last Exam lab | 29.7% ✎ | ▁▁▁▃▆▆▇██▆▁▃ beats 69/104 | 2.0% |
| Global-MMLU-Lite lab | 88.7% ✎ | ▂···▂·▂▂▂·██ beats 8/12 | 1.9% |
| CritPt lab | 5.4% ✎ | ▄···▄▄▄▄█▄▄▄ beats 4/9 | 1.7% |
| Icelandic LLM Leaderboard average (Miðeind, Sept 2026 suite) | 46.5% ✎ | ▁▁▃▂▂▄▁▃▄▄▄█ beats 17/38 | 1.6% |
| IFBench (aggregation unspecified) lab | 79.8% ✎ | ▃▇▂▄▅▅▂▄▄▆█▇ beats 74/80 | 1.3% |
$ where to run it
Through an API
api · Deep Infra · $1.72/M · $0.95 in / $4.05 out per 1M tokens · 1M context · deepinfra.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.
llama.cpp · recommended
version
- Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.macOS / Linux
curl -LsSf https://llama.app/install.sh | shWindowsirm https://llama.app/install.ps1 | iex - Run Inkling (975B-A41B) · needs ~585.5 GB · from huggingface.co/unslothchat in the terminal
llama cli -hf unsloth/inkling-GGUF:UD-Q4_K_XL - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf unsloth/inkling-GGUF:UD-Q4_K_XL
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run Inkling (975B-A41B) · needs ~585.5 GB · from huggingface.co/unsloth
- Or download it in the terminal
lms get https://huggingface.co/unsloth/inkling-GGUF@UD-Q4_K_XL -y - Then run lms ls to see its name, and lms chat followed by that name
Official weights and documentation: huggingface.co/thinkingmachines ↗
$ how the score was worked out · Epoch AI — internal evaluatio… 41%, thinkingmachines/Inkling-Smal… 33%, LMArena leaderboard dataset 22% and 2 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 Epoch AI — internal evaluations | 6 | 41% |
 thinkingmachines/Inkling-Small model card · lab-reported | 19 | 33% |
 LMArena leaderboard dataset | 3 | 22% |
 τ²/τ³-bench leaderboard: Sierra's own evaluation runs | 1 | 3% |
| Miðeind Icelandic LLM Leaderboard | 1 | 2% |
| prior · pull toward the middle | 0% | |
| weighted average | = 69.7 |
not counted (1)
| thinkingmachines/Inkling-Small model card | MMAU [correct] | 77.2% | fewer than three models have this test yet, or this source ran it for this model alone |
$ 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