From the source
Mistral AI released Voxtral 3, a speech model family with three transcription models, one text-to-speech model and a voice agent SDK built on LiveKit, all in public preview on the Mistral API.
- The family covers batch transcription and diarization, streaming ASR over WebSocket with end-of-turn detection, and text-to-speech, with all transcription models supporting diarization and context biasing across 50 languages.
- Mistral reports Voxtral Small Transcribe 3 at 2.1% AA-WER, third on Artificial Analysis, and best-in-chart diarization at 18.4% cpWER, ahead of Pyannote and MAI-Transcribe-2.
- The realtime model trades latency for accuracy, scoring 5.7% WER at 2,400 ms and 7.4% at 240 ms on FLEURS 35, against 4.2% for Gemini 3.5 Realtime Transcribe.
- Batch transcription costs $0.0035 to $0.004 per minute, Voxtral Speech 3 costs $13 per 1M characters, and the voice agent API costs $0.04/min with Mistral Small 4 or $0.08/min with any other reasoning model.
- The models run on Mistral's European, US and global endpoints, and OpenRouter lists all three new models with EU-hosted, zero-data-retention endpoints; no open weights were announced.