Voxtral TTS 3
Voxtral TTS 3 turns text into speech and supports both zero-shot and professional voice cloning. Mistral's October 2026 preview lists more than 35 languages and 50 dialects. The API uses the model identifier voxtral-mini-tts-3-0.
Preview
Separate voice control from response speed
The model page lists speech generation and voice cloning under /v1/audio/speech. It reports approximately 90 milliseconds to first audio for streaming. That vendor figure does not specify the total delay of a complete voice assistant.Voxtral TTS 3 model page
Evaluate the audio workflow
Practical guidance. Evaluate pronunciation, language switching and the stability of the selected voice over long passages. Measure time to the first audible word in the complete application. Emotion and pace controls are useful only if the chosen voice remains clear for the intended listener and content.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Launch date | |
| Access | API |
| Model identifiers | voxtral-mini-tts-3-0 |
| Initial release stage | Public preview.Voxtral TTS 3 model page |
| Controls | 34 emotion tags and two pace tagsVoxtral TTS 3 model page |
| Output | Streaming speech in five output formatsVoxtral TTS 3 model page |
| Access | Premier API preview; no public weights confirmedVoxtral TTS 3 model page |
Related models
- Voxtral Mini Transcribe Realtime 3
Mistral AI ยท Voxtral
A streaming transcription model with turn detection, speaker diarization and configurable latency over WebSocket.
Launch date PreviewAPI