Mistral releases Voxtral 3 speech models and a voice agent SDK

Mistral's 20 October 2026 post introduces Voxtral 3, a family of speech-to-text and text-to-speech models plus a voice agent SDK. Mistral says the SDK, built on LiveKit, runs listening, reasoning, tool calls, and speaking as one stream. A realtime transcription model listens, a general model such as Mistral Small reasons and calls tools, and a speech model replies. Mistral says each turn finishes in under a second, and the agent stops when the caller interrupts.

Mistral says Voxtral Small Transcribe 3 is state of the art on multilingual transcription and diarization, and that it outperforms Whisper, Scribe v2, and Gemini 3.5 Transcribe. The Mini transcription model has 4 billion parameters. Both support context biasing for domain terms. Voxtral Realtime Transcribe 3 is built for agents, with transcription Mistral describes as under 500 milliseconds in more than 35 languages, plus end-of-turn detection. Voxtral Mini TTS 3 covers more than 35 languages and more than 50 dialects. Pro Voice adapts a speaker from 30 minutes of audio.

On the API, Mistral lists Mini batch transcription at $0.0035 a minute, Small transcription at $0.004 a minute, realtime transcription at $0.008 a minute, and Voxtral Speech 3 at $13 per million characters for batch and realtime. A voice agent that reasons with Mistral Small 4 is $0.04 a minute. Any other reasoning model is $0.08 a minute. Mistral says the API is available today on European, US, and global endpoints. Running the full stack on-premises or in a private cloud means talking to Mistral.

Who it's for: teams building contact-center agents, captions, or a voice interface that has to stay on infrastructure they choose.

Fits: no ranked category. This is a voice stack. Best for Writing stays Claude, and the verdicts are unchanged.