Magistral is designed to make a model's intermediate reasoning available in the language of the task. In its June 2025 introduction, Mistral AI describes training for multistep reasoning across languages including French, Arabic and Simplified Chinese.
The intended benefit is that someone working through a complex question can inspect the proposed steps without switching to English. That is a language and interaction capability; the announcement's stronger claims about professional accuracy or regulatory compliance would require separate evidence. Displayed reasoning gives a user material to examine, rather than establishing that the conclusion is correct.
Terms in this piece · Glossary
reasoning model — A model trained to think — generating extended internal reasoning before answering — trading time and tokens for accuracy on hard problems.
