Voxtral 3
Mistral AI · v3 · depuis 2026-10-20 · 9× · vu le 22 oct. 2026
—
Momentum
Voxtral 3 is a model family from Mistral AI for speech applications, comprising speech-to-text (batch and realtime transcription), text-to-speech (Voxtral Speech 3), and a LiveKit-based voice agent SDK with tool calling. It was announced on October 20, 2026 and is available via Mistral's EU-hosted API, US/global endpoints, or self-deployment. Performance claims come mostly from Mistral's own evaluations.
Historique du momentum
24.07.22.10.
Fonctionnalités
| Real-Time Streaming | Yes: Voxtral Mini Transcribe Realtime 3 handles live streaming; streamed pipeline of transcription, reasoning and speech |
| Latency | Realtime transcription under 500 ms; TTS about 90 ms time-to-first-audio |
| License | Proprietary (Mistral Premier) per release overview for Voxtral TTS 3; no open weights for Voxtral 3 documented |
| Platform | Mistral API (EU, US and global endpoints) plus self-deployment on own infrastructure |
| Price | Mini Transcribe 3 (batch) $0.0035/min; Small Transcribe 3 $0.004/min; Realtime 3 $0.008/min; TTS (Speech 3) $13/1M characters |
| Release Date | October 20, 2026 |
| Languages | Transcription: 50+ languages; speech output: 35+ languages and 50+ dialects |
| Voice Cloning | Yes, zero-shot and professional voice cloning (Pro Voice with 30 min audio) per release overview for Voxtral TTS 3 |