Skip to main content

Magistral Medium 1

MistralOpen WeightApache 2.0 · Commercial OK

Description

Trained solely with reinforcement learning on top of Mistral Medium 3, Magistral Medium is a reasoning model that achieves strong performance on complex math and code tasks without relying on distillation from existing reasoning models. The training uses an RLVR framework with modifications to GRPO, enabling improved reasoning ability and multilingual consistency.

Release Date
2025-06-10
Parameters
24.0B
Context Length
262K
Modalities
image, text

Capability Radar

50
general
50
coding
50
reasoning
60
scienceest.
50
agents
10
multimodal

Science is estimated from LLM Stats science scores or reasoning when no dedicated science benchmarks are available.

Rankings

No ranking data available

Benchmark Scores (LLM Stats)

(LLM Stats (zeroeval))

General

Aider-Polyglot47.1%SR

Math

AIME 202473.6%SR
AIME 202564.9%SR

Reasoning

GPQANYU + Cohere + Anthropic (2023)70.8%SR
LiveCodeBench50.3%SR
Humanity's Last Exam9.0%SR

AA Evaluation Indices

(Artificial Analysis)

No AA evaluation data available

LLM Stats Category Scores

(LLM Stats (zeroeval))
Physics
70
Biology
70
Chemistry
70
Math
50
Reasoning
50
General
50
Code
50
Vision
10

Pricing

Input Price$2 / 1M tokens
Output Price$5 / 1M tokens
Blended Price (3:1)$2.75 / 1M tokens

Speed

No speed data available

Provider Price Ranking

Provider Price Ranking

1 providers

ProviderInputOutput
1MistralPRIMARY
$2
$5

Compare pricing across different API providers for this model.

External Sources