Mistral launches Voxtral 3 speech models and voice agent SDK
The family includes three transcription models and one speech model covering 50 languages; Voxtral Small Transcribe 3 ranks third on the Artificial Analysis speech-to-text leaderboard.
Mistral AI released the Voxtral 3 family and a LiveKit-based voice agent SDK on October 20. The lineup covers three transcription models and one speech model. Voxtral Small Transcribe 3 supports 50-plus languages with diarization and end-of-turn detection at 0.0040 dollars per minute; the 4B Voxtral Mini Transcribe 3 costs 0.0035 dollars per minute in batch; the realtime variant costs 0.008 dollars per minute with sub-500ms latency. Mistral claims Small Transcribe 3 beats Whisper, Scribe v2 and Gemini 3.5 Transcribe and ranks third on the Artificial Analysis speech-to-text leaderboard. Voxtral Mini TTS 3 covers 35-plus languages and 50-plus dialects at 13 dollars per million characters. The SDK chains transcription, reasoning and speech for sub-second turns over SIP and WebRTC. Licensing terms were not stated.