Ministral 3B

Compact Mistral model for edge, latency-sensitive, and cost-efficient workloads

Strongest at
Science
Better than 0% of the past year's models · one benchmark so far
Context window
128K
tokens in · 8.2K out

Capability profile

Where it lands among 91 models released in the past year. 50 is the average model.

This modelEvery model from the past year
Coding
No results yet
Writing
No results yet
Reading & documents
No results yet
Reasoning
No results yet
Math
No results yet
Science
25Better than 0% · 1 test
Visual reasoning
No results yet
Computer use
No results yet
Where these scores come from

Each capability combines the published benchmarks listed here, rescaled so 50 is the reference average and 15 points is one standard deviation. A hollow marker rests on a single benchmark.

The Ministral line

8 releases in the Ministral line.

A newer release is out: Ministral 3 8B 2512.

ECI STANDING#205 of 270
SCORE118.04106.31–121.89 CI
FRONTIER LAGLaunched 19.1 months behindScores about what GPT-4 (Mar 2023) did in March 2023.
SCORES LIKE THESE MODELSMistral NeMoMistral AI · Jul 2024118.65
Qwen2.5-7BAlibaba · Sep 2024118.54
GPT-3.5 Turbo (Nov 2023)OpenAI · Nov 2023118.53
Mixtral 8x7BMistral AI · Dec 2023118.44
Full Epoch Capabilities Index ranking ↗

Epoch benchmark boards

Best result on each board it has run. The tick marks the leader.

MATH level 5
14%#87 of 98

Mistral's past year

15 release days in the past year, about one every 24 days. The next is expected around Nov 16, 2026. Forecasts ↗

NovDecJanFebMarAprMayJunJulAugSepOctNov

Specifications

RELEASE DATEOct 16, 2024
WEIGHTSOpen weights
CONTEXT WINDOW128K
MAX OUTPUT8.2K
REASONINGNot listed
TOOL CALLINGNot listed
INPUTStext
OUTPUTStext

Models.dev record ↗

Specifications and prices from Models.dev; availability and limits may differ by provider. Benchmarks from Epoch AI (CC BY 4.0), LiveBench and the other sources named above.