Mistral has released Voxtral 3, a new family of speech-to-text and text-to-speech models paired with a voice agent SDK built for real-time, multi-step conversations across 50+ languages. The release targets contact centers, virtual assistants, and hands-free copilots that need to listen, reason, act, and speak in a single live pipeline.
What ships
- Voxtral Small Transcribe 3 — the speech-to-text model. Mistral calls it state-of-the-art for multilingual transcription, with speaker diarization (who said what), context-biasing via keyphrases to improve accuracy, and end-of-turn detection so agents can respond without delay.
- Voxtral Mini TTS 3 — the text-to-speech model. New voices adapt emotion, pace, and emphasis based on conversational context, with support for 35+ languages and 50+ dialects.
- Voice agent SDK — combines listening, reasoning, tool calling, and speaking into one real-time streaming pipeline. Agents can retrieve information, interact with business systems, and complete multi-step tasks during a live phone or web conversation.
- Pro Voice — a dedicated feature for highly accurate voice adaptation and cloning.
Sovereign by design
Like other Mistral models, the Voxtral 3 family is built for sovereign deployment. It is available through Mistral's EU-hosted API or for self-deployment on your own infrastructure, giving organizations control over where audio is processed, how data is handled, and whether conversations are retained.
Voxtral 3 is available today via the Mistral API on European, US, and global endpoints, and can be tested in the Studio playgrounds for voice agents, transcription, speech generation, and voice cloning.
Source: Mistral AI