Model

Mistral Small 3.2 (24B)

OPEN

Mistral AI · Mistral Small family · released Jun 20, 2025

Drop-in upgrade over 3.1: markedly better instruction following, roughly half the infinite-generation errors, more reliable function calling.

ReasoningCodingVisionFunction callingTool useAgentic

1075.6

Elo · rank #196

Parameters

24B

Active params

24B (dense)

Context

128K tokens

Architecture

Dense transformer (24B), vision-language, incremental update of Small 3.1

License

Apache 2.0

Languages

—

API price (in/out)

$0.08 / $0.2

Modalities

text · vision

Benchmark results

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

AI2DVision92.9%#15

best: Molmo 72B · 96.3%

Arena-Hard v2Human preference43.1%#8

best: Qwen3-Max · 86.1%

ChartQAVision87.4%#12

best: MiniMax-VL-01 · 91.7%

DocVQAVision94.9%#9

best: Qwen2-VL-72B · 96.5%

GPQA DiamondReasoning46.1%#202

best: GPT-5.6 · 94.6%

HumanEval+Coding92.9%#1

best: this model · 92.9%

MATH-500Math69.4%#93

best: GPT-5 · 99.4%

MathVistaVision67.1%#35

best: Seed 2.1 Pro · 90.7%

MBPP+Coding78.3%#7

best: Llama 3.1 405B · 88.6%

MMLU-ProKnowledge69.1%#94

best: Claude Fable 5 · 91.5%

MMLUKnowledge80.5%#86

best: OpenAI o3 · 92.9%

MMMUVision62.5%#72

best: Claude Fable 5 · 89.3%

SimpleQAKnowledge12.1%#33

best: GPT-4.5 · 62.5%

WildBenchHuman preference65.3%#1

best: this model · 65.3%

Run it locally

VRAM @ Q4

14.5 GB

VRAM @ FP16

55 GB

Fits on (Q4)

RTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

~70 tok/s on RTX 4090 (Q4, llama.cpp)

Quantizations

GGUF Q4 · AWQ · MLX

Mistral Small family

Elo progression across releases

Mistral Small 3 (24B)Jan 2025978.8 Mistral Small 3.1 (24B)Mar 20251051.4 Mistral Small 3.2 (24B)Jun 20251075.6

Resources

HFmistralai/Mistral-Small-3.2-24B-Instruct-2506 DOCSdocs.mistral.ai APIAvailable via Mistral AI API

API price $0.08/$0.2 · each benchmark row carries its own source badge (see methodology)