Inference
EN
Mistral

Magistral Medium

Magistral Medium is a reasoning model from Mistral in the Magistral family. 20 benchmarks count toward its index, in 7 categories.

  • Closed weights
  • Reasoning
  • Input: text
  • Released 2025-06
RankedEnough results for an index.
IndexAll categories, out of 100.33.7±4.1Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.—Median 67/s
First tokenSeconds to the first output token.—Median 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.40KMedian 262K
EloLMArena rating and rank.1255±6 #299Median 1422

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.3 counted30.6199 / 218
CodingCode writing and repair.3 counted39.2180 / 209
ReasoningLogic problems and puzzles.4 counted28.4212 / 227
MultimodalTasks with images and text.1 counted39.2—
KnowledgeFacts and expert knowledge.4 counted35.2224 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted12.5—
MathMath problems.4 counted41.8122 / 168

Results

20 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
AIMEMathAbove fitFar above this model's index.83.5% ±1.042 / 95
LiveCodeBenchCoding74.9% ±1.181 / 142
American Invitational Mathematics Examination 2024Math73.6%17 / 27
Critical Physics TasksReasoning0.3%137 / 214
Graduate-Level Google-Proof Q&AKnowledge70.8%92 / 141
Artificial Analysis Humanity's Last ExamKnowledge9.8%147 / 222
MATH-500 Problem SetMath94.3%18 / 27
Artificial Analysis GPQA DiamondKnowledge67.9%138 / 199
Terminal-Bench 1.0Agentic28.8% ±5.133 / 47
American Invitational Mathematics Examination 2025Math64.9%38 / 53
Artificial Analysis Omniscience AccuracyKnowledge20.3%128 / 164
GPQA Diamond · Vals AIKnowledge62.4% ±2.4106 / 135
MMMU ProMultimodal64.6% ±1.177 / 92
τ²-Bench Tool-Agent-User EvaluationAgentic23.1%125 / 149
LiveCodeBench v6Coding50.3%53 / 59
ARC-AGI-2 (semi-private)Reasoning0.0%120 / 131
MMLU ProKnowledge68.7% ±0.4125 / 135
ARC-AGI-1 (semi-private)Reasoning5.9%117 / 126
Artificial Analysis Long Context ReasoningReasoning0.0%204 / 217
Terminal-Bench 2.0 · Vals AIAgentic13.5% ±3.663 / 66
Artificial Analysis IFBenchInstruction25.1%140 / 146
IOI v1Coding0.7% ±0.459 / 61
MedQAHealthcare89.5% ±0.351 / 95
ARC-AGI-2 (public eval)Reasoning0.0%102 / 122
TaxEval v2Finance61.9% ±1.0121 / 143
CorpFin v2Finance47.4% ±1.0111 / 131
MortgageTaxFinance51.9% ±1.083 / 97
ARC-AGI-1 (public eval)Reasoning8.9%109 / 116
MGSMMultilingual74.6% ±2.571 / 74
LegalBenchLegal54.8% ±0.6143 / 149
LiveCodeBench v5Coding59.4%—

Sources

BenchLMCC BY-NC 4.0Vals AINo licence statedARC Prize FoundationTerms restrict reuse

History

5 releases
Index34.7 → 33.7
23 Oct 202612 Nov 2026

Questions

How good is Magistral Medium?
Its index is 33.7 out of 100. That is place 239 of 283 ranked models.
What is Magistral Medium best at?
Math: 41.8 out of 100, place 122 of 168.
What is the context window of Magistral Medium?
40K tokens.
Is Magistral Medium open source?
No. Its weights are not published.
When was Magistral Medium released?
June 2025.

Badge

Magistral Medium on Inference360: place 239 of 283, index 33.7
HTML
<a href="https://inference360.ai/models/magistral-medium"><img src="https://inference360.ai/badges/magistral-medium.svg" alt="Magistral Medium on Inference360: place 239 of 283, index 33.7" height="20"></a>
Markdown
[![Magistral Medium on Inference360: place 239 of 283, index 33.7](https://inference360.ai/badges/magistral-medium.svg)](https://inference360.ai/models/magistral-medium)

Compare with

Celeris-133.7o1-mini33.5Mistral Large 334.3Llama 4 Maverick34.4Magistral Small31.4

More from Mistral

All models
Magistral Small31.4Mistral Small 3.2 24BUnrankedDevstral Small28.2Mistral Medium 327.1Codestral 2508Unranked