Term
Magistral
Magistral is Mistral AIs first reasoning model from June 2025 — with a transparent, verifiable chain of thought and strength in European languages, partly as open weights.
Magistral — explained in more detail
Magistral is Mistral AIs first reasoning model and was released on June 10, 2025. Reasoning models generate an explicit chain of thought before the actual answer; Magistral emphasizes making that reasoning transparent and verifiable. A distinguishing feature is its pronounced multilinguality: the model is specifically designed for reasoning in European languages, whereas many competitors are optimized primarily for English.
Key facts
- Release: June 10, 2025.
- Variants: Magistral Small (24B parameters, open weights under the Apache 2.0 license, downloadable from Hugging Face) and Magistral Medium (a more powerful, API-only enterprise variant).
- Focus: transparent, verifiable reasoning and multilinguality.
- Benchmarks: Magistral Medium 73.6 percent on AIME2024 (90 percent with majority voting @64); Magistral Small 70.7 percent (83.3 percent @64).
Example / Practical use
Magistral suits tasks requiring multi-step inference — mathematical and logical problems, structured analysis and planning. The open Small variant can be self-hosted and fine-tuned, relevant for privacy-sensitive or European-language applications. The traceable chain of thought makes it easier to check results rather than trusting a black box.
Delimitation
Magistral is Mistrals entry into the reasoning class and competes with models such as OpenAIs o-series, DeepSeek R1 and the thinking modes of Claude and Gemini. The open Small variant differs from purely proprietary reasoning models through self-hosting capability; the Medium variant targets enterprise use via the API. Its particular focus on European languages sets Magistral apart from largely English-centric competitors.