[whatmodel.sh](../../README.md) / Ministral 3 8B
# Ministral 3 8B
Mistral · released 2025-12 · 8.92B params · apache-2.0
**score 52.9** ±0.8 · #318 of 732 · from 23 benchmark results across 9 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 8B among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆●▄▅▅▄▄▂▁ #223 of 313* 3.6% 1
chat & writing ▁▁▁▃▆●▆█▆▇▇▃ #293 of 402* 2.8% 1
math ▁▁▃▅█●▄█▄▂▁▁ #218 of 431 23% 4
agent tasks · #223 of 313 · 3.6% of its score
Beats [o1 (2024-12-17)](o1-2024-12-17.md), [o3-mini](o3-mini.md), [DeepSeek R1 (Jan '25)](deepseek-r1.md) here, though they rank higher overall. Trails [SmolLM3 3B](smollm3-3b.md), [Jamba 1.5 Mini](jamba-1-5-mini.md), [Mellum2 12B-A2.5B Instruct](mellum2-12b-a2-5b-instruct.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
ContractBench success rate (33 ob… 19.2% ▆··●▄▂▄▂▆▆█▄ beats 4/23 very hard · avg 13% 3.6% sharp
chat & writing · #293 of 402 · 2.8% of its score
Beats [Ministral 3 14B Reasoning 2512](ministral-3-14b-reasoning.md) here, though they rank higher overall. Trails [EXAONE 3.5 2.4B Instruct](exaone-3-5-2-4b.md), [LFM2 700M](lfm2-700m.md), [Flan-T5 XXL](flan-t5-xxl.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
Arena-Hard (version/judge unspeci… 50.9% ▃▃▂▅●▆▅█▅▅▂▃ beats 9/30 medium · avg 46% 2.8% at its level · small run
math · #218 of 431 · 23% of its score
Beats [MiniMax-M2.1](minimax-m2-1.md), [DeepSeek V3 (Dec '24)](deepseek-v3.md), [GPT-4o (2024-08-06)](gpt-4o-2024-08-06.md) here, though they rank higher overall. Trails [Granite 8B Code Base (4K)](granite-8b-code-base-4k.md), [Mixtral 8x7B v0.1 (base)](mixtral-8x7b-v0-1.md), [Tulu 2 DPO 70B](tulu-2-dpo-70b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
MATH (subset unspecified) 86.2% █▄··▄█●····▄ beats 6/7 easy · avg 80% 11% several runs
GSM8K 89.5% █····█··▄·▄● beats 6/7 very easy · avg 86% 6.5% sharp
PolyMath (Marco evaluation) 26.4% ▄··▄●█··▄··▄ beats 2/7 hard · avg 29% 3.0% at its level
MGSM (Marco evaluation) 68.7% ▄·▄·▄▄●·▄··█ beats 4/7 easy · avg 69% 2.5% at its level
other tests (15)
test result vs the others in its run difficulty counts why
Medmarks-V mean win rate (19 veri… 0.4442 ▁▄▃▃●▄█▃▅▁▁▁ beats 9/30 avg 0.418 13% sharp
GPQA Diamond 44.8% ▄···█·▄·█·●▄ beats 6/7 hard · avg 37% 12% sharp
MMLU 79.8% ▄···▄▄·▄█··● beats 6/7 easy · avg 74% 7.9% at its level
MMLU-Pro 63.9% ▄····█·▄█·●▄ beats 6/7 medium · avg 54% 6.0% at its level
Medmarks-V weighted mean win rate… 44.4% ▂▃▅▂●▃█▅█▅▃▆ beats 9/34 medium · avg 41% 5.7% blunt · at its level
EuroEval Life in the UK (English … 89.7% ▂·▂▂·▂·▃●·█▃ beats 6/14 easy · avg 80% 4.7% typical
MMLU-ProX-Lite (Marco) 43.3% █··▄··●··█·▄ beats 3/7 medium · avg 42% 4.5% at its level
GlobalMMLU (Marco evaluation) 55.4% ▃··▃▃··●···█ beats 3/7 medium · avg 56% 3.2% at its level
MMMLU (Marco evaluation) 56.4% ▄···▄·●·▄·▄█ beats 2/7 medium · avg 59% 2.8% at its level
EuroEval HellaSwag (English), val… 73.1% ▂··▂▂·▂●▆█·▄ beats 6/15 medium · avg 63% 2.8% at its level
KOPA-Bench (pass@1) 13.0% ●·█··▄··█▄·█ beats 0/8 very hard · avg 14% 2.4% typical
AGIEval (Marco suite unspecified) 52.4% ▄··▄█▄··▄··● beats 7/7 medium · avg 46% 2.2% blunt · at its level
SurgiQ surgical MCQ (zero-shot, l… 54.4% ▂···▂▂·▃▂●██ beats 6/17 medium · avg 53% 1.7% at its level
BFCL multi-turn (version not stat… 24.5% ▄●···▄·▄█▄▄▄ beats 1/8 hard · avg 26% 1.5% at its level
Vectara hallucination leaderboard… 78.3% ●▂▁▂▅█▂▃▄▃▁▁ beats 3/98 very easy · avg 90% 0.0% blunt
source results share of the score
Marco-MoE: Open Multilingual Mixture-of… 11 55% ▇▇▇▇▇▇▇
Medmarks: A Comprehensive Open-Source L… 2 19% ▇▇
EuroEval leaderboards 2 7% ▇
mistralai/Ministral-3-14B-Instruct-2512… 2 6% ▇
Multi-Step Tool-Calling over Korean Ope… 2 4% ▇
ContractBench: Can LLM Agents Preserve … 1 4% ▇
mistralai/Ministral-3-8B-Instruct-2512 … 1 3% ▇
SurgiQ: A Large-Scale Multi-Domain Benc… 1 2% ▇
Vectara hallucination leaderboard 1 0% ▇
prior: pull toward the middle 0% ▇
weighted average = 52.9
Licences: Marco-MoE: Open Multilingual Mixture-of-Expert Language Models with Efficient Upcycling ([CC BY 4.0](../../licenses/cards-final-NOTICE.txt)) · Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks ([CC BY 4.0](../../licenses/paper-2605.01417v1-CC-BY-4.0.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE)) · mistralai/Ministral-3-14B-Instruct-2512 model card ([Apache-2.0](../../licenses/cards-gap-NOTICE.txt)) · Multi-Step Tool-Calling over Korean Open Public APIs: A Benchmark and a Data-Synthesis Recipe ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · ContractBench: Can LLM Agents Preserve Observation Contracts? ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · mistralai/Ministral-3-8B-Instruct-2512 model card ([Apache-2.0](../../licenses/cards-gap-NOTICE.txt)) · SurgiQ: A Large-Scale Multi-Domain Benchmark for Evaluating Surgical Understanding in Large Language Models ([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))
## How to run it
### Through an API
api (Mistral) · $0.15/M blended · $0.15 in / $0.15 out per 1M tokens · [docs.mistral.ai](https://docs.mistral.ai/models/ministral-3-8b-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-8b-2512. Uncached input; output includes generated reasoning where billed. Batch, cache storage, tools and other modalities excluded. · [context source](https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/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 8B (Q4_K_M, needs ~5.2 GB, from [huggingface.co/mistralai](https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512-GGUF/tree/0102285ad796bd99af90f58de616092e5630e970))
```sh
llama cli -hf mistralai/Ministral-3-8B-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-3-8B-Instruct-2512-GGUF:Q4_K_M
```
Other versions: Q8_0 (~9 GB) `llama cli -hf mistralai/Ministral-3-8B-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 8B (Q4_K_M, needs ~5.2 GB, from [huggingface.co/mistralai](https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512-GGUF/tree/0102285ad796bd99af90f58de616092e5630e970))
```sh
lms get https://huggingface.co/mistralai/Ministral-3-8B-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 (~9 GB) `lms get https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512-GGUF@Q8_0 -y` · 4bit (~4.8 GB, Mac only) `lms get https://huggingface.co/alexgusevski/Ministral-3-8B-Instruct-2512-q4-mlx -y` · 8bit (~9 GB, Mac only) `lms get https://huggingface.co/alexgusevski/Ministral-3-8B-Instruct-2512-q8-mlx -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 8B (Q4_K_M, needs ~5.85 GB, from [ollama.com](https://ollama.com/library/ministral-3/tags))
```sh
ollama run ministral-3:8b
```
Other versions: Q8_0 (~11.2 GB) `ollama run ministral-3:8b-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 8B (4bit, needs ~4.8 GB, Mac only, from [huggingface.co/alexgusevski](https://huggingface.co/alexgusevski/Ministral-3-8B-Instruct-2512-q4-mlx/tree/952bf8a6613ba41585cf9e640fec854034877c8e))
```sh
mlx_lm.chat --model alexgusevski/Ministral-3-8B-Instruct-2512-q4-mlx
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model alexgusevski/Ministral-3-8B-Instruct-2512-q4-mlx
```
Other versions: 8bit (~9 GB, Mac only) `mlx_lm.chat --model alexgusevski/Ministral-3-8B-Instruct-2512-q8-mlx`
Model repository and documentation: [huggingface.co/mistralai](https://huggingface.co/mistralai/Ministral-3-8B-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 67.76 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-3-8b) · [JSON](https://whatmodel.sh/api?format=json&all=1)