Command Palette

Search for a command to run...

Mistral Medium 3

Mistral

Vision ModelProprietary

Context

Release Date
May 07, 2025
Window
128k

PricingPer 1M tokens

Input
$0.4
Output
$2
Blended 3:1
$0.8

Capabilities

Speed
78 t/s
Input
Output
Reasoning tokens

Latency

TTFT
0.36 ms
500 token response
6.80 s

Benchmarks

Intelligence
●●○○○
Math
●●●○○
Coding
●●○○○
MMLU Pro
76.0%
GPQA
57.8%
HLE
4.3%
SciCode
33.1%
AIME
44.0%
MATH 500
90.7%
LiveCodeBench
40.0%
HumanEval
89.8%

Mistral Medium 3 is a frontier-class language model that matches or beats larger competitors on coding, STEM, and multimodal benchmarks while costing 8× less ($0.40/M input, $2/M output). It runs in any cloud or on-prem setup with as few as four GPUs, giving you near-Claude-Sonnet accuracy without heavyweight infrastructure.

For developers, that means high-quality code generation, data analysis, and multilingual chat in a budget-friendly package. You can fully fine-tune, keep the model inside your VPC for compliance, and drop it into existing tools to power secure chatbots, enterprise search, or domain-specific agents without paying proprietary prices.