Mistral AI introduced Voxtral 3 around October 20, 2026. The post carries no visible publication date, so we use the date in its benchmark footnotes. Mistral describes it as a family of speech-to-text and text-to-speech models plus a voice agent SDK built on LiveKit that runs listening, reasoning, tool calling and speaking as one real-time pipeline.
What to know
- Transcription: Mistral lists Voxtral Small Transcribe 3 at $0.0040 per minute, Voxtral Mini Transcribe 3 (batch) at $0.0035 per minute and Voxtral Mini Transcribe Realtime 3 at $0.008 per minute, with sub-500ms latency for the realtime model.
- Text-to-speech: Voxtral Mini TTS 3 is listed at $13 per 1M characters, with controls for emotion, pacing and emphasis.
- Voice agents: the agent API is listed at $0.04 per minute with Mistral Small 4 for reasoning, or $0.08 per minute with a larger model.
- Languages: Mistral says transcription covers 50+ languages, including the 24 official EU languages, and text-to-speech covers 35+ languages.
- Voice cloning: Mistral says it needs 30+ minutes of audio.
- Where: Mistral’s EU-hosted API, with US and global endpoints, or self-deployment. The page does not state a license.
- Benchmarks: Mistral says Voxtral Small Transcribe 3 beats Whisper, Scribe v2 and Gemini 3.5 Transcribe on multilingual transcription. These are the company’s own claims, not independent tests.
Why it matters
Voice agents are a fast-growing way to use AI, and per-minute pricing makes costs easy to estimate. If you build one, our guide to writing voice agent prompts covers how to instruct the model. For creative voices, compare with ElevenLabs Eleven v4 and see Mistral and the model tracker.
Only clone a voice with the speaker’s permission.
Source: Mistral AI. We summarise in our own words; check the source for the full details.