Mistral AI launches full Voxtral 3 real-time voice agent stack

Voxtral 3 is a set of models for real-time voice agents in three parts: the Voxtral Small Transcribe 3 speech transcription model, the Voxtral Mini TTS 3 text-to-speech model, and a LiveKit-based voice agent SDK. Together they chain listening, reasoning, tool calling, and speaking into a single streaming pipeline. Transcription covers 50+ languages, while speech synthesis supports 35+ languages and 50 dialects with emotion adaptation. Agents can connect to external systems mid-conversation, such as booking appointments or checking bills. On deployment, Mistral says the solution can run on your own infrastructure or be called via an EU-hosted API, keeping data within your control; transcription pricing starts at $0.0035 per minute, with details in the official docs. For product teams building real-time voice support or voice assistants, the value is a packaged mix of transcription, synthesis, and tool calling with optional private deployment; for developers handling EU user voice data, the EU-hosted API is a new compliance option.