Mistral AI launches Magistral Small & Medium, its first series of reasoning models

Mistral AI has announced Magistral, its first reasoning model, available in both open and enterprise editions. This release introduces Magistral Small, a 24 billion parameter model with open weights under the Apache 2.0 license for self-deployment, and Magistral Medium, designed for enterprise applications.

Magistral is tailored for professional domains by supporting verifiable reasoning processes. The model can handle structured calculations, programmatic logic, decision trees, and rule-based systems, making it suitable for a range of enterprise workflows. Magistral is optimized for multilingual reasoning. It performs especially well in English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese.

Magistral Medium also introduces Think mode and Flash Answers in Le Chat. These features deliver speeds up to ten times faster than many competing enterprise models. Mistral Small is now available from Hugging Face, while Mistral Medium can be found on Amazon SageMaker, and soon on IBM WatsonX, Azure AI and Google Cloud Marketplace.