Model

Ministral 3B

CLOSED

Mistral AI · Ministral family · released Oct 16, 2024

Edge-class 3B model for on-device/low-latency use; API-only, weights never released; retired from platform as of late 2025.

ReasoningCodingVisionFunction callingTool useAgentic

423.7

Elo · rank #325

Parameters

Active params

3B (dense)

Context

128K tokens

Architecture

Dense Transformer, 36 layers

License

Mistral Commercial License (proprietary; API-only, weights withheld)

Languages

—

API price (in/out)

$0.04 / $0.04

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AGIEvalReasoning42.1%#29

best: OLMo 3-Think 32B · 88.2%

ARC-ChallengeReasoning64.2%#69

best: Llama 3.1 405B · 96.9%

Arena-HardHuman preference64.3%#38

best: Qwen3-235B-A22B (Non-Thinking) · 96.1%

GSM8KMath50.9%#133

best: Llama 3.1 405B · 96.8%

HumanEvalCoding77.4%#60

best: Claude Opus 4.5 · 99.4%

MATH-500Math51.7%#126

best: GPT-5 · 99.4%

MBPPCoding67.7%#44

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLUKnowledge60.9%#202

best: OpenAI o3 · 92.9%

MT-BenchHuman preference8.1#44

best: Hunyuan-Large (A52B) · 9.4

TriviaQAKnowledge56.7%#36

best: Sarvam-1 (2B) · 90.6%

WildBenchHuman preference36.3%#9

best: Mistral Small 3.2 (24B) · 65.3%

WinoGrandeReasoning72.7%#83

best: PaLM 2 · 90.9%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$0.04

Output / M tok

$0.04

Ministral family

Elo progression across releases

Ministral 3BOct 2024423.7 Ministral 8BOct 2024553.7

Resources

DOCSdocs.mistral.ai/models/overview

API price $0.04/$0.04 · each benchmark row carries its own source badge (see methodology)