Mistral’s first Voxtral TTS model generates speech in nine languages

Mistral introduced Voxtral TTS on 23 March, adding speech generation to the Voxtral family. The four-billion-parameter model supports nine languages and voice adaptation, with the company emphasising expressive delivery and low time to first audio. Access opened through Mistral Studio and the API.

The launch positions the model for voice agents and enterprise audio workflows, with pricing starting at $0.016 per thousand characters. Character-based pricing differs from token-based language-model billing, so an application’s speech cost depends on the text actually spoken. Mistral’s claims about naturalness and emotional range come from its own evaluation. Teams building a recognisable synthetic voice also need permission for any reference recordings, consistency checks and a clear account of when listeners are hearing generated speech.