Benchmark Almanac

Ministral 3 3B Reasoning

Ministral 3 3B Reasoning.

Mistral AI Released December 2nd, 2025 Family Ministral 3 The lab's model card Compare with others
15
Benchmarks with a score
0
Ranked first, out of five or more models
0
In the top three
Mistral AI Dec 2025

Ministral 3 3B Reasoning

Open weights Small

No Epoch Capabilities Index score yet
  • Angry Tweets 48.7% #99 /149
  • FoQA 45.6% #57 /91
  • GermanQuAD 58.9% #29 /88

How does it do against its rivals?

Ministral 3 3B Reasoning beats Gemma 3 270M on 12 of the 12 benchmarks they share. Gemma 3n E4B does best against it, ahead on 8 of 8.

  1. Gemma 3 270M 12 ahead, 0 behind Compare
  2. LFM2-1.2B 10 ahead, 2 behind Compare
  3. gpt-oss-20b 4 ahead, 9 behind Compare
  4. Qwen3-4B-Thinking-2507 3 ahead, 9 behind Compare
  5. Olmo 3.1 32B Instruct 2 ahead, 10 behind Compare
  6. Gemma 3n E4B 0 ahead, 8 behind Compare

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
GermanQuAD 58.9% #29 of 88 Official leaderboard
EuroEval Faroese Grammatical Error Correction 0.97% #30 of 71 Official leaderboard
EuroEval ERRNews 29.1 #37 of 76 Official leaderboard
Dutch Book Reviews Dataset 88.5% #54 of 105 Official leaderboard
OrangeSum 35.1% #19 of 34 Official leaderboard
BLiMP-NL 26.9% #12 of 21 Official leaderboard
Dutch CoLA 26.9% #12 of 21 Official leaderboard
FoQA 45.6% #57 of 91 Official leaderboard
Angry Tweets 48.7% #99 of 149 Official leaderboard
DaLA 13.1% #18 of 25 Official leaderboard
FoSent 27.4% #113 of 147 Official leaderboard
GuiaCat 43.9% #29 of 35 Official leaderboard
AgentIF (2026-10 setting) 53.8% #24 of 24 Official leaderboard

Professional work

BenchmarkScoreRankSource
KorMedMCQA-Mixed 50.7% #23 of 33 Official leaderboard
KorMedMCQA-V 45.2% #26 of 33 Official leaderboard

Who measured it?

Independent runs cover all 15 of its benchmarks.

  • An independent run only 15
Official leaderboards scored it on 15.

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards