Voxtral 3: four Mistral speech models and a voice agent SDK
What is it?
Voxtral 3 is a family of speech models: three transcription (STT) models (Small Transcribe 3, the 4B Mini Transcribe 3, and Mini Transcribe Realtime 3) and one text-to-speech model, Mini TTS 3. A voice agent SDK built on LiveKit ties listening, reasoning and speaking into one stream. Transcription covers 50+ languages, with speaker diarization and hints for domain terms.
Why does it matter?
It's cheap: batch transcription runs $0.0035 to $0.004 per minute, realtime $0.008 per minute, and TTS $13 per million characters. Mistral says Small Transcribe 3 beats Whisper, Scribe v2 and Gemini 3.5 Transcribe on multilingual FLEURSVendor claim, with realtime latency under 500msVendor claim. If speech costs dominate your product, as in call centers or meeting notes, it's worth a comparison. EU endpoints and on-prem deployment help teams with data residency rules.
Maturity
PreviewYou can call it on the Mistral API today, but several outlets describe it as a public preview. The announcement doesn't say whether weights will be released.
How to track it
- Independent evals: check whether Voxtral 3 shows up on the Artificial Analysis speech leaderboards.
- Weights: watch the mistralai org on Hugging Face for Voxtral 3 checkpoints.
- SDK: follow GitHub releases for the voice agent SDK and any new Korean voice presets.