Magistral

The first reasoning model by Mistral AI

Visit website →

About

Magistral is Mistral AI's first model built specifically for reasoning. It works through problems step by step, showing a transparent chain of thought, and is designed to handle domain-specific reasoning across many languages rather than only English.

It comes in two versions: Magistral Small, a 24-billion-parameter model released with open weights that can be run and fine-tuned freely, and Magistral Medium, a larger enterprise model available through Mistral's API and platform.

It is aimed at developers and organisations that need strong multi-step reasoning for analysis, research, and structured tasks, and that value being able to inspect the model's reasoning and, for the open version, host it themselves.

AI-assisted draft, human-reviewed before publishing — see how we choose & review tools.

Why we picked it

Magistral stood out for the open-weight Small version specifically — a 24B reasoning model you can actually run and inspect yourself, plus multilingual chain-of-thought rather than English-only, is worth naming against closed reasoning models.