Mistral Small 3.2 24B (2506)

The latest iteration of Mistral Small, version 3.2 (2506) brings enhanced performance and capabilities. With 24 billion parameters, this model delivers state-of-the-art results across text generation tasks with improved efficiency.

Added Apr 15, 2025

Model weights

Pricing

Auto routing · per 1M tokens

Input: $0.20
Output: $0.40

Compare provider prices

Specifications

Context window: 128K
Max output: 131.1K
Parameters: 24B
Avg output (7d): 285 tokens; Longer than 23% of models

Benchmarks

Artificial Analysis

LMArena

Sourced from Artificial Analysis.

Intelligence Index

8.2

Better than 30% of models compared

Coding Index

12.5

Better than 12% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

0.3%

Better than 6% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

0 Elo

Better than 3% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

-212 Elo

Better than 5% of models compared

Document reasoning

AA-LCR v1.1

Long context reasoning with updated grading

20.3%

Better than 22% of models compared

Reasoning

HLE

Humanity's Last Exam

4.3%

Better than 18% of models compared

IFBench

Instruction-following benchmark

33.5%

Better than 23% of models compared

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

0.0%

Better than 16% of models compared

SciCode

Python programming for scientific computing

28.6%

Better than 9% of models compared

LiveCodeBench

Contamination-free coding benchmark

27.5%

Better than 30% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

27.0%

Better than 29% of models compared

AIME

American Invitational Mathematics Examination

32.3%

Better than 59% of models compared

Math-500

Diverse mathematical problem solving benchmark

88.3%

Better than 59% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

68.1%

Better than 32% of models compared

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

50.5%

Better than 24% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

6.8%

Better than 37% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

29.5%

Better than 37% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation