Voxtral Mini TTS 3
mistral · Released Oct 2026
Conditional. Voxtral Mini TTS 3 is Mistral's text-to-speech model on the Mistral API, billed at $13 per 1M characters, with zero-shot and professional voice cloning, streaming and emotion tags. Mistral's docs and launch post now agree on 35+ languages and 50+ dialects. The deciding caveat is that the model is still a public preview and its quality rankings are Mistral's own charts. Suits voice-agent builders who will audition voices in the Mistral playground and test their own language before committing.
Is it right for you?
Good for
- Voice agents that need streaming speech; the docs page lists about 90 ms time-to-first-audio (Mistral-stated)
- Branded or custom voices: the docs page lists zero-shot and professional voice cloning and the launch post describes cloning from 30+ minutes of sampled audio plus Pro Voice adaptation
- Expressive delivery control: the docs page lists 34 emotion tags and 2 pace tags, and the post lists emotion, pacing and emphasis
- Cost-aware speech at a stated $13 per 1M characters, which Mistral says is below OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash
Not good for
- Plans that depend on a specific language list: Mistral's docs now say 35+ languages and 50+ dialects, matching the launch post, but we have not tested coverage for your language
- Self-hosting from open weights; the pages we read state no open-weights release for Voxtral 3 (an earlier Voxtral TTS from March 2026 is described on the news index as open-weights, which is a different release); the launch post says to talk to Mistral's team for on-premises or private cloud deployment
- Unverified quality claims: the charts on Voice Code Bench, pronunciation control and multilingual quality are Mistral's own
How it performs by task
Voice agent speech output
Streaming at about 90 ms time-to-first-audio per the docs page, paired with Mistral's voice-agent SDK and Voxtral transcription; Mistral-stated, untested by us
Voice cloning
Zero-shot and professional cloning are documented; consent and restriction terms are not detailed on the pages we read
Multilingual speech
Mistral's docs and launch post now both say 35+ languages and 50+ dialects; the quality claims are Mistral's own and untested by us
Pricing
Input
$13 / 1M characters
Output
Billed per character (single rate)
Context
Per-character billing; docs: Public Preview
Verdict history
Sources
Verification log
- Pricing· No changes
Automated agent
- Pricing· No changes
Automated agent
- Profile· Updated
Automated agent
Docs page and models overview now state 35+ languages and 50+ dialects (stored: docs said 9 languages). Price $13 per 1M characters unchanged.
- Pricing· No changes
Automated agent