Current outcome
Mistral released Voxtral 3, a modular speech stack with transcription models, a synthesis model, and a LiveKit-based Voice Agent SDK, supporting 50+ languages for transcription and real-time voice agents; the model remains closed-source.
Progress timeline
2 material updates- #01
Voxtral 3
Mistral released Voxtral 3, an upgraded speech model offering transcription, diarization, and speech-generation capabilities, though it remains closed-source.
Source evidence: ilreb
- #02
Mistral发布Voxtral 3：把语音识别、推理和工具调用接成实时Agent
Added details that Voxtral 3 is a modular speech stack with three transcription models, a synthesis model, and a LiveKit-based Voice Agent SDK; transcription supports 50+ languages, synthesis supports 35+ languages and 50+ dialects, with API pricing disclosed.
State after update: Mistral released Voxtral 3, a modular speech stack with transcription models, a synthesis model, and a LiveKit-based Voice Agent SDK, supporting 50+ languages for transcription and real-time voice agents; the model remains closed-source.
Source evidence: theblockbeats