Mistral AI deployed Magistral, its first family of reasoning models, released in June 2025, which are designed to work through complex problems step-by-step in multiple languages, making their reasoning process visible and auditable to users.

Magistral is available in two versions: Magistral Small (24 billion parameters, open-source under Apache 2.0 on Hugging Face) and Magistral Medium (enterprise version, accessible via Le Chat and Mistral's API). The models use chain-of-thought reasoning and are fine-tuned for structured calculations, logic, decision trees, and rule-based systems across eight languages. Magistral Medium is available on Amazon SageMaker and planned for IBM Watsonx, Microsoft Azure AI, and Google Cloud Marketplace.