Voxtral TTS is Mistral's text-to-speech model. According to its model page, version 26.03 (API id voxtral-mini-tts-2603) was released on March 23, 2026, supports 9 languages and zero-shot voice cloning, streams audio with about 90 ms time to first audio, and is served from the /v1/audio/speech endpoint. Mistral has since listed a newer model, Voxtral TTS 3, described below.
Newer model
Mistral's models overview now lists Voxtral TTS 3 (voxtral-mini-tts-3-0). Its model page shows a release date of October 20, 2026, Public Preview status, 35+ languages and 50+ dialects, zero-shot and professional voice cloning, about 90 ms time to first audio with streaming, and a price of $13 per million characters. The pages do not describe 26.03 as deprecated, and they do not say whether the new model replaces it.
Key features
- Zero-shot voice cloning: clone a voice from a short voice prompt. The page says voice prompts do not need a transcript.
- Low latency: about 90 ms time to first audio, with streaming.
- Multilingual: 9 languages in version 26.03. The page does not list them.
- License: CC BY-NC 4.0, a non-commercial license.
Model specs
| Spec | Voxtral TTS 26.03 |
|---|---|
| API model id | voxtral-mini-tts-2603 |
| Release date | March 23, 2026 |
| Languages | 9 |
| Time to first audio | About 90 ms, streaming |
| License | CC BY-NC 4.0 |
| Endpoint | /v1/audio/speech |
Pricing
The model page, checked on October 29, 2026, lists $16 per million characters for version 26.03. Confirm on Mistral's pricing page before budgeting.
Use cases
- Voice agents that need fast first audio.
- Narration and voice-over with a cloned voice you have the right to use.
- Pairing with speech recognition, such as Voxtral Mini Transcribe 2, for a full voice loop.
Getting started
- Create a Mistral API key and call
/v1/audio/speechwith the Voxtral TTS model. - Provide a short voice prompt for voice cloning, and enable streaming for low latency.
- Check the license terms before any commercial use, since the listed license is CC BY-NC 4.0.
FAQ
How many languages does Voxtral TTS support?
Version 26.03 supports 9 languages. The newer Voxtral TTS 3 lists 35+ languages and 50+ dialects.
How much does it cost?
$16 per million characters for 26.03, and $13 per million for Voxtral TTS 3, as listed on October 29, 2026.
Does it clone voices?
Yes. Version 26.03 offers zero-shot voice cloning, and Voxtral TTS 3 adds professional voice cloning.
Can I use it commercially?
The page lists CC BY-NC 4.0, which is non-commercial. Check the terms that apply to your access route.
Alternatives
- Voxtral Mini Transcribe 2: Mistral's transcription model.
- OpenAI tts-1: OpenAI's earlier text-to-speech model.
- ElevenLabs Turbo v2.5: a low-latency speech model from ElevenLabs.
Conclusion
Voxtral TTS combines streaming speech, voice cloning and a published per-character price. See the model page for details, and consider Voxtral TTS 3 if you need broader language coverage.
Sources
- Websitehttps://docs.mistral.ai/models
- Social Mediahttps://x.com/MistralAI
Comments
No comments yet. Be the first to comment!