Voxtral TTS 3
Audio modelCard summarizedState-of-the-art expressiveness and instruction following across 35+ languages and 50+ dialects
Source: Voxtral 3 launch post and Voxtral Text to Speech docs (no model card published)
Where to call it
- Mistral API (Mistral Studio)
voxtral-mini-tts-3-0
Good for
- Voice agents and contact centers (with the realtime variant)
- Reading out names, numbers and rare words, with phoneme hints
- Brand voices via zero-shot or professional cloning
Overview
Voxtral TTS 3
State-of-the-art expressiveness and instruction following across 35+ languages and 50+ dialects
- Max length
- —
- Languages
- 35+ languages, 50+ dialects
- Modes
- text-to-speech, zero-shot voice cloning
- Released
- 20 Oct 2026
Voxtral TTS 3 (called Voxtral Mini TTS 3 and Voxtral Speech 3 in the launch post) is Mistral's text-to-speech model for the Voxtral 3 family. It speaks with preset, zero-shot cloned or professionally cloned voices, takes inline emotion and pace tags and phoneme hints, and streams audio. A realtime variant serves live conversations at the same $13 per 1M characters. Mistral says it reads context to pick emotion and voice style, and positions it as cheaper than OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash. launch post; Text to Speech docs
voxtral-mini-tts-3-0- Typical price
- $0.013 / 1K chars
- Languages
- 35+ languages, 50+ dialects (launch post, Languages page); models page says 9
- Weights
- Closed
- Released
- 20 Oct 2026
- text-to-speech
- zero-shot voice cloning
- professional voice cloning
- emotion and pace tags
- custom pronunciation (phoneme hints)
- streaming (SSE)