Mistral releases Voxtral 3: three speech models, TTS and a voice agent SDK

Mistral AI launched Voxtral 3, a family of three transcription models (Small, 4B-parameter Mini and Realtime), the Voxtral Speech 3 text-to-speech model and a LiveKit-based voice agent SDK. All four models are in public preview on the Mistral API, with batch transcription from $0.0035/min and TTS at $13 per 1M characters.

• Small Transcribe 3 scores 2.1% AA-WER, ranking third on Artificial Analysis
• Diarization: 18.4% cpWER for Small vs 19.0% for Pyannote and 20.3% for MAI-Transcribe-2
• Realtime model hits 7.4% WER on FLEURS 35 at 240 ms latency, vs 4.2% for Gemini 3.5 Realtime
• Batch transcription covers 50 languages, including 24 official EU languages