Mistral introduces Voxtral 3 voice models and a LiveKit-based agent SDK

Mistral announced Voxtral 3 on Oct. 20, describing Transcribe and Speech models for its new voice agents API. The company also introduced a voice agent SDK built on LiveKit. In Mistral’s account, Transcribe listens, a reasoning model thinks and calls tools, and Speech delivers the spoken reply in a streamed pipeline.

Mistral says Voxtral Small Transcribe 3 handles multilingual transcription, diarization and context-biasing. The company says it can also detect when someone has finished speaking so an agent can respond. For spoken output, Mistral says Voxtral Mini TTS 3 offers expressive speech in 35-plus languages and 50-plus dialects. Mistral says tags such as can set emotion and pace line by line.