whatmodel.sh / Mistral Large 3 (675B-A41B)

Mistral Large 3 (675B-A41B)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Mistral Large 3 (675B-A41B) is an open-weights language model from Mistral, released December 2025. Its published size is 675B parameters (41B active per token) and a 262k-token context window. Licence: apache-2.0.

61.7±1.3#181 of 719 LLMsfrom 15 benchmark results across 7 sources

$ how it compares

The LLMs ranked either side of Mistral Large 3 (675B-A41B), out of 828 with a score, then more from Mistral.

#modelscorewhereRAMreleased
211Qwen3-VL 32B Thinking Alibaba62±1api $0.28/M +local~22GB2025-10
212Gemini 2.0 Flash-Lite (001) Google62±2comparison only2025-02
213Mistral Large 3 (675B-A41B) Mistral62±1api $0.75/M +local~108GB2025-12
214Apriel 1.6 15B Thinker ServiceNow62±2local~10GB2025-11
215Gemini 2.5 Flash-Lite Preview (06-17) Google62±2comparison only2025-06
more from Mistral
79Mistral Large 4 (public preview) Mistral71±2api $2.06/M2026-10
174Mistral Medium 3.5 (128B) Mistral65±1api $3.00/M +local~42GB2026-04
180Mistral Medium 3.1 (2508) Mistral64±2comparison only2025-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #209 of 300 * · 3.9% of its score · 1 test

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 (Jan '25) here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Granite 4.1 3B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Thinkingbox-bench pass@1 (507 stateful business-workflow agent tasks, N=20)4.7% ✎▅··▃·▃·█▃▅██ beats 0/15very hard · avg 4.9%▇▇ 3.9%typical
coding▁▁▁▃▆██▇▆▄▃▁ #170 of 458 * · 26% of its score · 1 test

Beats Qwen3-Coder-Next (80B-A3B), Qwen3-VL 235B-A22B Instruct, Qwen3.5 9B here, though they rank higher overall. Trails QwQ-32B-Preview, Ministral 3 8B Reasoning 2512, Phi-4-reasoning 14B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1229.8 ✎▁▂▄▃▅█▄▇▆▄▁▁ beats 8/119avg 981▇▇▇▇▇▇▇▇▇▇ 26%sharp
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #96 of 393 * · 20% of its score · 2 tests

Beats LongCat 2.0 (1.6T-A48B), Inkling Small (266B-A12B), GPT-5.4 nano here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
LMArena (text)1413.9 ✎▁▁▁▂▃▃▃▆▆▇█▃ beats 262/379avg 1268▇▇▇▇▇▇ 15%blunt
Arena-Hard (version/judge unspecified) lab66.7% ✎▃▃▂▅▃▆▅█▅▅▂▃ beats 17/30medium · avg 50%▇▇ 4.9%at its level · small run
vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #81 of 147 * · 15% of its score · 2 tests

Beats DeepSeek V4 Flash Vision (Exp), GPT-5 mini, o1 (2024-12-17) here, though they rank higher overall. Trails GPT-4.1 mini, Gemini 2.5 Flash here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena vision (style controlled)1200.6 ✎▁▁▂▂▂▂▄▅▅▆██ beats 71/145avg 1068▇▇▇▇ 10%blunt
MMMU Pro lab54.0% ✎▂▃▁▃▆▇▃▇▇██▅ beats 16/52hard · avg 42%▇▇ 5.1%at its level · small run
other tests9 not in a task · 34% of its score
testresultvs the others in its rundifficultycountswhy
GPQA Diamond lab64.1% ✎▁▁▁▁▂▄▅▅███▄ beats 126/310hard · avg 43%▇▇▇▇ 10%sharp · at its level · small run
MMLU-Pro lab80.9% ✎▁▁·▁▁▁▃▂▄▅█▁ beats 130/188medium · avg 60%▇▇ 5.8%small run
LongSocialBench (full-context, 4-option MCQ over discussion threads)42.0% ✎▆▂·▄█·▂▂·▄▄▂ beats 9/16very hard · avg 36%▇▇ 4.5%blunt
IFBench (aggregation unspecified) lab37.8% ✎▃▇▂▄▅▅▂▄▄▆█▇ beats 25/80hard · avg 31%▇▇ 4.5%at its level · small run
Icelandic LLM Leaderboard average (Miðeind, Sept 2026 suite)25.7% ✎▁▁▃▂▂▄▁▃▄▄▄█ beats 9/38very hard · avg 14%▇▇ 4.3%at its level
WMT25 Terminology Translation (term success rate)91.2% ✎▂·▂··▂·▄▂·█▆ beats 11/12medium · avg 64%▇ 3.1%noisy
WMT24++78.3% ✎█·▃·▅▅▃···▃▃ beats 4/10easy · avg 71%▇ 1.3%blunt
WMT26 General MT (GEMBA-ESA, LLM judge)81.6% ✎▄·▄▄▄·▄██▄█▄ beats 11/12easy · avg 65%▇ 1.1%blunt · at its level
1 more test
testresultvs the others in its rundifficultycountswhy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)85.5% ✎▁▂▁▃▅█▂▃▄▃▁▁ beats 12/98very easy · avg 90%▇ 0.0%blunt

$ where to run it

Through an API

api · Mistral AI · $0.75/M · $0.5 in / $1.5 out per 1M tokens · 262k context · docs.mistral.ai ↗ · price sourceUSD per 1M text tokens, standard uncached input and output. Product: La Plateforme chat completions. API model: mistral-large-2512. Settings: service tier=standard, batch=no, cache=uncached input. Additional/excluded charges: Tools, cache storage, priority service and other modalities excluded.. Currency: USD.

llama.cpp · recommended

version

  1. 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 | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Mistral Large 3 (675B-A41B) · needs ~405.5 GB · from huggingface.co/unsloth
    chat in the terminal
    llama cli -hf unsloth/Mistral-Large-3-675B-Instruct-2512-GGUF:UD-Q4_K_XL
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf unsloth/Mistral-Large-3-675B-Instruct-2512-GGUF:UD-Q4_K_XL

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Mistral Large 3 (675B-A41B) · needs ~405.5 GB · from huggingface.co/unsloth

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/unsloth/Mistral-Large-3-675B-Instruct-2512-GGUF@UD-Q4_K_XL -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Weights and documentation: huggingface.co/mistralai ↗

$ how the score was worked out · LMArena leaderboard dataset 51%, mistralai/Mistral-Small-4-119… 30%, Cohere — North Small Translat… 5% and 4 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

sourceresultsshare of the scorelicence
LMArena leaderboard dataset3▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 51%CC BY 4.0
mistralai/Mistral-Small-4-119B-2603 model card · lab-reported5▇▇▇▇▇▇▇▇▇ 30%Apache-2.0
Cohere — North Small Translate (arXiv:2609.13916v1)3▇▇ 5%CC BY 4.0
LongSocialBench: Do Long-Context LLMs Understand Online Discussion Threads? (arXiv:2610.04118v1)1▇ 4%CC BY 4.0
Miðeind Icelandic LLM Leaderboard1▇ 4%CC BY 4.0
One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business Workflows (arXiv:2608.19741v4)1▇ 4%CC BY 4.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)1▇ 0%Apache-2.0
prior · pull toward the middle▇ 1%
weighted average= 61.7

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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