[whatmodel.sh](../../README.md) / Ministral 3 3B # Ministral 3 3B Mistral · released 2025-12 · 3.85B params · apache-2.0 **score 46.1** ±0.8 · #453 of 732 · from 33 benchmark results across 8 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks Ministral 3 3B among every model on that task or test, worst to best. 
task                vs the others          rank  counts  tests
agent tasks         ▁▁●▆█▄▅▅▄▄▂▁   #296 of 313*    1.7%      1
chat & writing      ▁▁▁▃●▄▆█▆▇▇▃   #336 of 402*    3.6%      2
math                ▁▁▃▅●▆▄█▄▂▁▁    #275 of 431     28%      4
vision & documents  ▁▁▂●▁▃▄▂▆▇█▄    #136 of 153     10%      9
agent tasks · #296 of 313 · 1.7% of its score Beats [GLM-4 9B Chat](glm-4-9b-chat.md), [DBRX Instruct](dbrx-instruct.md), [QwQ-32B-Preview](qwq-32b-preview.md) here, though they rank higher overall. Trails [Granite 4.0 H 350M](granite-4-0-h-350m.md), [ChatGLM3-6B](chatglm3-6b.md), [Qwen3 0.6B](qwen3-0-6b.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty           counts  why
ContractBench success rate (33 ob…    6.1%  ●··▆▄▂▄▂▆▆█▄ beats 0/23   very hard · avg 13%    1.7%  sharp
chat & writing · #336 of 402 · 3.6% of its score Beats [OpenThinker3 7B](openthinker3-7b.md), [Code Llama 70B Instruct](codellama-70b-instruct.md), [Gemma 4 E2B](gemma-4-e2b.md) here, though they rank higher overall. Trails [LFM2 350M](lfm2-350m.md), [Gemma 3 1B](gemma-3-1b.md), [Qwen3 0.6B](qwen3-0-6b.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty        counts  why
Arena-Hard (version/judge unspeci…   30.5%  ▃●▂▅▃▆▅█▅▅▂▃ beats 3/30   medium · avg 46%    2.4%  at its level · small run
IFEval (aggregation unspecified)     72.5%  ▄···▄●▄····▄ beats 2/5    easy · avg 78%      1.2%  at its level
math · #275 of 431 · 28% of its score Beats [GPT-4 Turbo (2024-04-09)](gpt-4-turbo-2024-04-09.md), [Molmo 2 8B](molmo-2-8b.md), [Llama 3.3 70B Instruct](llama-3-3-70b.md) here, though they rank higher overall. Trails [DeepSeek R1 Distill Qwen 1.5B](deepseek-r1-distill-qwen-1-5b.md), [Gemma 7B (base)](gemma-7b-base.md), [CodeGemma 7B (v1.0)](codegemma-7b.md) here, though they rank lower overall. 
test                         result  vs the others in its run  difficulty           counts  why
MATH (subset unspecified)     70.2%  ●▄··▄█▄····▄ beats 0/7    easy · avg 80%          13%  at its level · several runs
GSM8K                         79.0%  ●····█··▄·▄█ beats 1/7    very easy · avg 86%     10%  sharp
MGSM (Marco evaluation)       47.0%  ●·▄·▄▄▄·▄··█ beats 0/7    easy · avg 69%         2.6%  at its level
PolyMath (Marco evaluation)   16.3%  ●··▄██··▄··▄ beats 0/7    hard · avg 29%         2.2%  typical
vision & documents · #136 of 153 · 10% of its score Beats [Gemma 4 E2B](gemma-4-e2b.md) here, though they rank higher overall. 
test                           result  vs the others in its run  difficulty        counts  why
CharXiv descriptive questions   76.6%  █·▄·······▄● beats 5/5    easy · avg 78%      3.2%  at its level
MMStar                          53.1%  ▄▄▄●·······▄ beats 4/5    medium · avg 61%    2.4%  at its level
MMBench-EN dev v1.1             69.2%  ●·····▃··▃·▃ beats 1/5    easy · avg 79%      1.3%  blunt · at its level
HallusionBench                  65.2%  ▄▄·▄·····●·▄ beats 3/5    easy · avg 66%      1.1%  blunt · at its level
RealWorldQA                     58.3%  ▄···●··▄▄··▄ beats 2/5    easy · avg 68%      0.6%  blunt · at its level
CountBench                      73.7%  ▄··▄··●··▄·▄ beats 3/5    easy · avg 84%      0.6%  noisy
OCRBench                        73.5%  █··█●█·█···█ beats 2/5    easy · avg 81%      0.4%  blunt
AI2D test                       74.1%  █···●▄···▄·▄ beats 2/5    easy · avg 81%      0.3%  blunt · at its level
ChartQA                         79.1%  ▄······▄··●█ beats 3/5    easy · avg 79%      0.2%  blunt · noisy
other tests (13) 
test                                result  vs the others in its run   difficulty           counts  why
MMLU                                 69.8%  ▄···●▄·▄█··█ beats 1/7     easy · avg 74%          12%  at its level · several runs
Medmarks-V mean win rate (19 veri…  0.3927  ▁▄●▃▇▄█▃▅▁▁▁ beats 4/30    avg 0.418               12%  sharp
MMLU-Pro                             49.5%  ▄····●·▄█·▄▄ beats 2/7     medium · avg 54%       8.4%  at its level · several runs
GPQA Diamond                         31.6%  ▄···●·▄·█·▄▄ beats 1/7     hard · avg 37%         6.0%  sharp
Medmarks-V weighted mean win rate…   39.3%  ▂▃●▂▆▃█▅█▅▃▆ beats 4/34    medium · avg 41%       4.8%  blunt · at its level
MMLU-ProX-Lite (Marco)               26.7%  ●··▄··█··█·▄ beats 0/7     medium · avg 42%       3.7%  at its level
GlobalMMLU (Marco evaluation)        38.4%  ●··▃▃··▅···█ beats 0/7     medium · avg 56%       2.6%  at its level
MMMLU (Marco evaluation)             39.4%  ●···▄·█·▄·▄█ beats 0/7     medium · avg 59%       2.4%  at its level
AGIEval (Marco suite unspecified)    44.7%  ▄··▄●▄··▄··█ beats 3/7     medium · avg 46%       2.0%  blunt · at its level
EuroEval HellaSwag (English), tes…   69.8%  ▃·▁▁▂▁▂▅●▇▄▁ beats 50/100  medium · avg 67%       1.2%  sharp · at its level · noisy
Multi-IF (aggregation unspecified)   47.0%  ▄·▄··●▄····▄ beats 3/5     medium · avg 57%       1.1%  at its level
EuroEval Life in the UK (English …   81.5%  ▁▁▁▁▁▁▁▂▄●▄▃ beats 72/146  easy · avg 82%         0.9%  at its level · noisy
Vectara hallucination leaderboard…   75.8%  ●▂▁▂▅█▂▃▄▃▁▁ beats 0/98    very easy · avg 90%    0.0%  blunt
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's 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 is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
Marco-MoE: Open Multilingual Mixture-of…       11   50% ▇▇▇▇▇▇
CohereLabs/North-Micro-Vision-Instruct …       13   19% ▇▇
Medmarks: A Comprehensive Open-Source L…        2   17% ▇▇
mistralai/Ministral-3-14B-Instruct-2512…        2    6% ▇
mistralai/Ministral-3-8B-Instruct-2512 …        1    4% ▇
EuroEval leaderboards                           2    2% ▇
ContractBench: Can LLM Agents Preserve …        1    2% ▇
Vectara hallucination leaderboard               1    0% ▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 46.1
Licences: Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling ([CC BY 4.0](../../licenses/cards-final-NOTICE.txt)) · CohereLabs/North-Micro-Vision-Instruct model card ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)) · Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks ([CC BY 4.0](../../licenses/paper-2605.01417v1-CC-BY-4.0.txt)) · mistralai/Ministral-3-14B-Instruct-2512 model card ([Apache-2.0](../../licenses/cards-gap-NOTICE.txt)) · mistralai/Ministral-3-8B-Instruct-2512 model card ([Apache-2.0](../../licenses/cards-gap-NOTICE.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE)) · ContractBench: Can LLM Agents Preserve Observation Contracts? ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Vectara hallucination leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0))
Not counted (1) 
source                  test                       score  why
mistralai/Ministral-3…  MATH (subset unspecified)  83.0%  only model in this source's run
## How to run it ### Through an API api (Mistral) · $0.10/M blended · $0.1 in / $0.1 out per 1M tokens · [docs.mistral.ai](https://docs.mistral.ai/models/ministral-3-3b-25-12) [price source](https://github.com/BerriAI/litellm/blob/c19ce71bcdad234ca87a7af10cee45375ce811a8/model_prices_and_context_window.json): USD per 1M text tokens. Mistral standard synchronous API; model ministral-3b-2512. Uncached input; output includes generated reasoning where billed. Batch, cache storage, tools and other modalities excluded. · [context source](https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/config.json): Official Instruct-2512 model configuration; card corroborates 256K. For 3B this replaces a missing value using the first-party record, not LiteLLM’s conflicting 131,072 entry. ### llama.cpp (recommended) 1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md) ```sh # macOS / Linux curl -LsSf https://llama.app/install.sh | sh # Windows irm https://llama.app/install.ps1 | iex ``` 2. Run Ministral 3 3B (Q4_K_M, needs ~2.1 GB, from [huggingface.co/mistralai](https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512-GGUF/tree/eb599d408350ea2bb60452cb86be7c7b2fc28227)) ```sh llama cli -hf mistralai/Ministral-3-3B-Instruct-2512-GGUF:Q4_K_M ``` 3. Or start a local server and open the address it prints (http://localhost:9931) ```sh llama serve -hf mistralai/Ministral-3-3B-Instruct-2512-GGUF:Q4_K_M ``` Other versions: Q8_0 (~3.7 GB) `llama cli -hf mistralai/Ministral-3-3B-Instruct-2512-GGUF:Q8_0` ### LM Studio 1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download) 2. Run Ministral 3 3B (Q4_K_M, needs ~2.1 GB, from [huggingface.co/mistralai](https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512-GGUF/tree/eb599d408350ea2bb60452cb86be7c7b2fc28227)) ```sh lms get https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512-GGUF@Q4_K_M -y ``` 3. Then run lms ls to see its name, and lms chat followed by that name Other versions: Q8_0 (~3.7 GB) `lms get https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512-GGUF@Q8_0 -y` · 4bit (~1.9 GB, Mac only) `lms get https://huggingface.co/Oscilla/Ministral-3-3B-Instruct-2512-mlx-4Bit -y` · 8bit (~3.6 GB, Mac only) `lms get https://huggingface.co/Oscilla/Ministral-3-3B-Instruct-2512-mlx-8Bit -y` ### Ollama 1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download) ```sh # Linux curl -fsSL https://ollama.com/install.sh | sh ``` 2. Run Ministral 3 3B (Q4_K_M, needs ~2.81 GB, from [ollama.com](https://ollama.com/library/ministral-3/tags)) ```sh ollama run ministral-3:3b ``` Other versions: Q8_0 (~5.12 GB) `ollama run ministral-3:3b-instruct-2512-q8_0` ### MLX (Mac) 1. Install MLX (Mac). Apple silicon Macs only, text only. ```sh # with uv uv tool install mlx-lm # or pip install mlx-lm ``` 2. Run Ministral 3 3B (4bit, needs ~1.9 GB, Mac only, from [huggingface.co/Oscilla](https://huggingface.co/Oscilla/Ministral-3-3B-Instruct-2512-mlx-4Bit/tree/10889ee5bad71785905c2fb444439e81823cdd7a)) ```sh mlx_lm.chat --model Oscilla/Ministral-3-3B-Instruct-2512-mlx-4Bit ``` 3. Or start an OpenAI-compatible server at http://localhost:8080/v1 ```sh mlx_lm.server --model Oscilla/Ministral-3-3B-Instruct-2512-mlx-4Bit ``` Other versions: 8bit (~3.6 GB, Mac only) `mlx_lm.chat --model Oscilla/Ministral-3-3B-Instruct-2512-mlx-8Bit` Model repository and documentation: [huggingface.co/mistralai](https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512) ## Published speed measurements Reference results for the hardware, engine and settings below. These are not verified speeds of the downloads above or predictions for your machine. 
hardware                      precision         speed  engine and settings
Apple M5 Max (40-core GPU),…  q4_k_m     127.64 tok/s  ollama (version not recorded) · l…
Each speed links to its source, which has the full settings. [interactive version](https://whatmodel.sh/model/ministral-3-3b) · [JSON](https://whatmodel.sh/api?format=json&all=1)