The LLM benchmark aggregator.

Ministral 8B

Mistral · 8B

Strongest in Factual grounding (#35 of 104), weakest in Human preference (#269 of 354). Above par in 2 of 6 scopes. Among the models it meets almost everywhere, it finishes behind GPT-4.1 and Gemini 2.5 Flash-Lite and ahead of DeepSeek R1 and NVIDIA Nemotron 3 Nano 30B A3B.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Safety52−9.4#130/3271/3

Factual grounding56.4−14.3#35/1041/1

Agents41.9−24.4#265/2981/6

Tool use30.6−43.4#77/811/1

Human preference41.1−27.7#269/3541/1

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Ministral 8B, left for the other.

Llama 3.2 1B Instructwon 5 · lost 0

Llama 3.1 8B Instructwon 4 · lost 1

Aya Expanse 8Bwon 3 · lost 1

Llama 3.2 3B Instructwon 3 · lost 2

Granite 3.1 8B Instructwon 3 · lost 2

NVIDIA Nemotron 3 Nano 30B A3Bwon 3 · lost 2

DeepSeek R1won 3 · lost 2

Snowflake Arctic Instructwon 2 · lost 2

Llama 3.1 Nemotron Ultra 253B v1won 2 · lost 2

GLM-4.7 Flashwon 2 · lost 2

MiniMax M2.1won 2 · lost 2

MiniMax M2.5won 2 · lost 2

Qwen3.5 Flashwon 2 · lost 2

Grok 4 Fastwon 2 · lost 2

Claude Opus 4.1won 2 · lost 2

Gemma 4 31Bwon 2 · lost 2

GLM-5won 2 · lost 2

Gemini 3 Flashwon 2 · lost 2

Claude Opus 4.6won 2 · lost 2

Llama 4 Scout Instructwon 3 · lost 3

Mercury 2won 2 · lost 3

Gemma 3 27B ITwon 2 · lost 3

DeepSeek V3won 2 · lost 3

Mistral Medium 3.1won 2 · lost 3

Mistral Large 3won 2 · lost 3

Gemini 3.1 Flash-Litewon 2 · lost 3

GLM-4.7won 2 · lost 3

Kimi K2.5won 2 · lost 3

Kimi K2.6won 2 · lost 3

DeepSeek V4 Prowon 2 · lost 3

Claude Sonnet 4.6won 2 · lost 3

GPT-6 Astrawon 2 · lost 3

Gemini 3.1 Prowon 2 · lost 3

Gemini 4 Argonwon 2 · lost 3

Command Awon 2 · lost 4

Gemini 2.5 Flashwon 2 · lost 4

Kimi K2won 2 · lost 4

GLM-4.6won 2 · lost 4

Grok 4.1 Fastwon 2 · lost 4

Claude Opus 4.5won 2 · lost 4

Gemini 3 Prowon 2 · lost 4

Qwen3 8Bwon 1 · lost 3

Aya Expanse 32Bwon 1 · lost 3

Mistral Small 3 24B Instructwon 1 · lost 3

Command R+won 1 · lost 3

GPT OSS 20Bwon 1 · lost 3

NVIDIA Nemotron 3 Super 120B A12Bwon 1 · lost 3

O3 Miniwon 1 · lost 3

GLM-4.5 Airwon 1 · lost 3

Qwen3.5 27Bwon 1 · lost 3

Grok 3won 1 · lost 3

DeepSeek V3.1won 1 · lost 3

Mistral Large 4won 1 · lost 3

NVIDIA Nemotron 3 Ultra 550B A55Bwon 1 · lost 3

Claude Opus 4won 1 · lost 3

MiniMax M3won 1 · lost 3

DeepSeek V4 Flashwon 1 · lost 3

Grok 4.3won 1 · lost 3

Inklingwon 1 · lost 3

MiMo V2.6 Flashwon 1 · lost 3

MiMo V2.5 Prowon 1 · lost 3

Qwen3.7 Maxwon 1 · lost 3

GLM-5.3 Flashwon 1 · lost 3

GLM-5.2won 1 · lost 3

Gemini 3.5 Flashwon 1 · lost 3

Gemini 3.6 Flashwon 1 · lost 3

GPT-4owon 1 · lost 4

GPT OSS 120Bwon 1 · lost 4

Trinity Largewon 1 · lost 4

Qwen3 Next 80B A3B Thinkingwon 1 · lost 4

Qwen3 30B A3B Instructwon 1 · lost 4

Qwen3.5 35B A3Bwon 1 · lost 4

Claude Sonnet 4won 1 · lost 4

Grok 4won 1 · lost 4

MiniMax M2.7won 1 · lost 4

Qwen3.5 122B A10Bwon 1 · lost 4

GPT-5won 1 · lost 4

Gemma 4 26B A4Bwon 1 · lost 4

Gemini 2.5 Prowon 1 · lost 4

GPT-5.1won 1 · lost 4

GPT-5.5won 1 · lost 4

GPT-5.6 Solwon 1 · lost 4

Claude Opus 4.7won 1 · lost 4

Claude Opus 5.5won 1 · lost 4

Gemma 3 4B ITwon 1 · lost 5

Llama 3.3 70B Instructwon 1 · lost 5

Llama 4 Maverick Instructwon 1 · lost 5

GPT-5 nanowon 1 · lost 5

Gemma 3 12B ITwon 1 · lost 5

GPT-5 miniwon 1 · lost 5

O4 Miniwon 1 · lost 5

Claude Haiku 4.5won 1 · lost 5

Qwen3 235B A22B Instructwon 1 · lost 5

DeepSeek V3.2won 1 · lost 5

GPT-5.2won 1 · lost 5

Claude Sonnet 4.5won 1 · lost 5

Qwen3 14Bwon 0 · lost 4

Qwen3 4B Instructwon 0 · lost 4

Amazon Nova Litewon 0 · lost 4

GPT-4won 0 · lost 4

Granite 4.0 H Smallwon 0 · lost 4

GPT-4o miniwon 0 · lost 4

Claude 3.5 Haikuwon 0 · lost 4

Mistral Small 3.2won 0 · lost 4

Mistral Medium 3won 0 · lost 4

Claude 3.7 Sonnetwon 0 · lost 4

GPT-6 Lunawon 0 · lost 4

Qwen3.5 397B A17Bwon 0 · lost 4

Grok 4.7won 0 · lost 4

GPT-5.6 Lunawon 0 · lost 4

Hy3won 0 · lost 4

Grok 4.6won 0 · lost 4

Claude Sonnet 5won 0 · lost 4

GLM-5.1won 0 · lost 4

Grok 4.5won 0 · lost 4

GPT-5.6 Terrawon 0 · lost 4

MiMo V2.6 Prowon 0 · lost 4

Claude Opus 4.8won 0 · lost 4

Qwen3.8 Maxwon 0 · lost 4

Claude Opus 5won 0 · lost 4

Muse Spark 1.1won 0 · lost 4

Gemini 3.8 Flashwon 0 · lost 4

GPT-4.1 nanowon 0 · lost 5

GPT-4.1 miniwon 0 · lost 5

GPT-5.4 nanowon 0 · lost 5

O3won 0 · lost 5

GPT-5.4 miniwon 0 · lost 5

GPT-6 Solwon 0 · lost 5

DeepSeek V4.1 Flashwon 0 · lost 5

GPT-5.4won 0 · lost 5

Amazon Nova Microwon 0 · lost 6

Phi 4won 0 · lost 6

Amazon Nova Prowon 0 · lost 6

Mistral Large 2.1won 0 · lost 6

Amazon Nova 2 Litewon 0 · lost 6

Qwen3 32Bwon 0 · lost 6

Gemini 2.5 Flash-Litewon 0 · lost 6

GPT-4.1won 0 · lost 6

§ 3 · Sources

Where the numbers come from

3 publications, 3 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-11-12

LMArena Text 1238

BFCL ↗1 measureread 2026-11-12

BFCL v4 11.1%

Vectara Hallucination Leaderboard ↗1 measureread 2026-11-12

Vectara · Factual consistency 92.6%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=ministral-8b)](https://publicai.io/model-index/m/ministral-8b)