Inference
EN
Mistral

Magistral Small

Magistral Small is a reasoning model from Mistral in the Magistral family. 18 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2025-06
RankedEnough results for an index.
IndexAll categories, out of 100.31.4±4.3Median 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.128KMedian 262K
EloLMArena rating and rank.—Median 1422

Run locally

23.57B parameters
16-bit
50.5 GB
8-bit
27.3 GB
4-bit
16.0 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy fits in 16 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.2 counted24.8215 / 218
CodingCode writing and repair.2 counted36.3189 / 209
ReasoningLogic problems and puzzles.4 counted28.2213 / 227
MultimodalTasks with images and text.1 counted40.0—
KnowledgeFacts and expert knowledge.4 counted31.0236 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted12.2—
MathMath problems.4 counted41.3124 / 168

Results

18 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
AIMEMathAbove fitFar above this model's index.80.7% ±1.247 / 95
MATH-500 Problem SetMath95.9%15 / 27
LiveCodeBenchCoding72.1% ±1.183 / 142
American Invitational Mathematics Examination 2024Math70.7%18 / 27
Graduate-Level Google-Proof Q&AKnowledge68.2%97 / 141
Artificial Analysis Humanity's Last ExamKnowledge7.6%155 / 222
Critical Physics TasksReasoning0.0%150 / 214
Artificial Analysis GPQA DiamondKnowledge64.1%146 / 199
American Invitational Mathematics Examination 2025Math62.8%40 / 53
τ²-Bench Tool-Agent-User EvaluationAgentic26.6%116 / 149
MMMU ProMultimodal65.2% ±1.176 / 92
GPQA Diamond · Vals AIKnowledge58.3% ±2.5112 / 135
Artificial Analysis Omniscience AccuracyKnowledge18.0%141 / 164
LiveCodeBench v6Coding47.4%54 / 59
ARC-AGI-2 (semi-private)Reasoning0.0%120 / 131
Terminal-Bench 1.0Agentic15.0% ±4.044 / 47
Artificial Analysis Long Context ReasoningReasoning0.0%204 / 217
ARC-AGI-1 (semi-private)Reasoning5.0%121 / 126
Artificial Analysis IFBenchInstruction24.8%141 / 146
MMLU ProKnowledge62.1% ±0.5132 / 135
MortgageTaxFinance62.1% ±1.057 / 97
MedQAHealthcare82.4% ±0.366 / 95
ARC-AGI-2 (public eval)Reasoning0.0%102 / 122
MGSMMultilingual86.3% ±2.162 / 74
TaxEval v2Finance60.3% ±1.0128 / 143
CorpFin v2Finance44.0% ±1.0119 / 131
ARC-AGI-1 (public eval)Reasoning8.6%110 / 116
LegalBenchLegal40.0% ±0.5148 / 149
LiveCodeBench v5Coding55.8%—

Sources

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

History

5 releases
Index32.3 → 31.4
23 Oct 202612 Nov 2026

Questions

How good is Magistral Small?
Its index is 31.4 out of 100. That is place 247 of 283 ranked models.
What is Magistral Small best at?
Math: 41.3 out of 100, place 124 of 168.
What is the context window of Magistral Small?
128K tokens.
Is Magistral Small open source?
Yes. Its weights are published.
When was Magistral Small released?
June 2025.

Badge

Magistral Small on Inference360: place 247 of 283, index 31.4
HTML
<a href="https://inference360.ai/models/magistral-small"><img src="https://inference360.ai/badges/magistral-small.svg" alt="Magistral Small on Inference360: place 247 of 283, index 31.4" height="20"></a>
Markdown
[![Magistral Small on Inference360: place 247 of 283, index 31.4](https://inference360.ai/badges/magistral-small.svg)](https://inference360.ai/models/magistral-small)

Compare with

LFM2.5-8B-A1B30.6Claude 3.5 Sonnet32.2Exaone 4.0 32B32.4Sarvam 30B30.3Magistral Medium33.7

More from Mistral

All models
Magistral Medium33.7Mistral Small 3.2 24BUnrankedDevstral Small28.2Mistral Medium 327.1Codestral 2508Unranked