Magistral Medium 1

Mistral开源权重Apache 2.0 · 商用许可

描述

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.

发布日期

2025-06-10

参数规模

24.0B

上下文长度

128K

支持模态

text

能力雷达图

general

coding

reasoning

science

agents

multimodal

排行榜排名

领域	#排名	分数	来源
代码能力榜	474	24.0	AA
通用能力榜	480	29.0	AA
科学能力	367	41.0	AA

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

General

Aider-Polyglot

47.1%自报

Math

AIME 2024

73.6%自报

AIME 2025

64.9%自报

Reasoning

GPQANYU + Cohere + Anthropic (2023)

70.8%自报

LiveCodeBench

50.3%自报

Humanity's Last Exam

9.0%自报

AA 评测指数

(Artificial Analysis)

Math 500(OpenAI (2024), subset of Hendrycks et al. MATH (2021))

91.7

Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))

75.3

Aime(MAA (Mathematical Association of America))

70.0

Gpqa(NYU + Cohere + Anthropic (2023))

67.9

Livecodebench(UC Berkeley + MIT + Cornell (2024))

52.7

Aime 25(MAA (Mathematical Association of America))

40.3

Math Index(Artificial Analysis)

40.3

Ifbench(Google Research (2023))

25.1

Tau2(Sierra + U Toronto + Vector Institute (2025))

23.1

Hle(Center for AI Safety + Scale AI (2025))

9.8

Intelligence Index(Artificial Analysis)

9.1

Terminalbench Hard(Stanford × Laude Institute (2026))

9.1

Lcr(Artificial Analysis)

0.0

LLM Stats 分类评分

(LLM Stats (zeroeval))

Physics

Biology

Chemistry

Math

Reasoning

General

Code

Vision

定价

输入价格免费

输出价格免费

混合价格(3:1)免费

速度

Tokens/秒0.0

首Token延迟0.00s

首回答延迟0.00s

供应商价格排行

1 个供应商

供应商输入输出

1Pioneer

比较该模型在不同 API 供应商之间的定价。

外部链接

LLM Stats Artificial Analysis