Mistral AI launches Voxtral 3, a full voice agent stack that listens, thinks, and talks in real time

A voice agent must hear you, decide what to say and reply fast enough to feel like a conversation. Mistral's Voxtral 3 bundles a speech-to-text model, a text-to-speech model and a developer kit built on LiveKit into one streaming pipeline, transcribes more than 50 languages, and lets agents call outside tools mid-call, on your own servers if you prefer.

Summary written with AI by HowTechMoves, based on reporting by The Decoder.