The system combines listening, reasoning, tool calls and speech generation in a streaming pipeline.
What happened
The release covers speech-to-text and text-to-speech models, multilingual conversations and connections to an organization's own tools and data. Mistral says the family is available through its EU-hosted API or for deployment on customers' infrastructure.
Why it matters
A voice assistant needs more than transcription accuracy. It must handle interruptions, retain the user's intent and carry out the right action while keeping the conversation understandable and responsive.
The bigger picture
Our take. Test complete conversations, including corrections and ambiguous requests. A fluent voice is useful, but the decisive result is whether the system understood the request and executed the intended workflow with clear confirmation.
Source checked November 18, 2026. Company claims are attributed; paragraphs marked “Our take” are editorial interpretation.