Magistral Small
Mistral's first open reasoning model: 24B parameters of traceable chain-of-thought under Apache 2.0.
About
Magistral is Mistral AI's first dedicated reasoning model family, launched June 10, 2025 at London Tech Week. It ships in two variants: Magistral Small, a 24B-parameter open model under Apache 2.0 on Hugging Face, and Magistral Medium, an enterprise-grade closed model on Le Chat and cloud marketplaces.
Both models reason step by step with transparent, traceable thought chains, and both reason natively in eight languages: English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese. Mistral reports 73.6% on AIME 2024 for Medium and 70.7% for Small, rising to 90% and 83.3% with majority voting. A later 1.2 update added image analysis to Small.
The honest caveats: the benchmark figures are vendor-reported, and only the Small variant is open, with Medium remaining proprietary. It is positioned for finance, healthcare, government, and legal work where auditable reasoning matters.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 87; Trending score 82/100 from researcher assessment.