Command Palette

Search for a command to run...

Mistral Small 3.2

Mistral

Vision ModelOpen

Context

Release Date
Jun 20, 2025
Window
128k

PricingPer 1M tokens

Input
$0.1
Output
$0.3
Blended 3:1
$0.15

Capabilities

Speed
153 t/s
Input
Output
Reasoning tokens

Latency

TTFT
0.26 ms
500 token response
3.52 s

Benchmarks

Intelligence
●●○○○
Math
●●○○○
Coding
○○○○
MMLU Pro
68.0%
GPQA
51.0%
HLE
4.3%
SciCode
26.0%
AIME
32.0%
MATH 500
88.0%
LiveCodeBench
28.0%
HumanEval
85.0%

Mistral-Small-3.2-24B-Instruct is a 24 B-parameter, Apache-2.0 LLM that tightens instruction following, halves repetition loops, and adds sturdier function calling compared to v3.1. It scores 65 % on WildBench v2 and ~80 % MMLU while handling text-and-image prompts in 24 languages.

Developers can self-host it with vLLM or Transformers on ~55 GB of GPU VRAM, skipping API fees and lock-in. Strong function-calling and vision support make it a practical open-source base for chatbots, agent pipelines, and multimodal apps.