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

Mistral AI has launched Voxtral 3, a family of language models and supporting components aimed at real-time voice agents. The release bundles three parts: Voxtral Small Transcribe 3 for speech-to-text, Voxtral Mini TTS 3 for text-to-speech, and a voice agent SDK built on LiveKit. Mistral describes them as forming a single streaming pipeline in which an agent listens, reasons, calls tools, and speaks.
On coverage, transcription handles more than 50 languages, while the speech output supports over 35 languages and 50 dialects, with what Mistral calls emotional adaptation. The agents can reach into external systems mid-conversation, for example to book appointments or look up invoices, which is the piece that turns a voice interface into something that can actually complete tasks rather than just converse.
Deployment is offered either on the customer's own infrastructure or through EU-hosted APIs, a positioning that keeps data handling under the customer's control. Pricing starts at $0.0035 per minute for transcription. Mistral points to its documentation for further detail.
This suggests Mistral is aiming at the full loop rather than selling the speech components piecemeal; LiveKit as the SDK foundation is a widely used open-source framework for real-time audio and video, so teams may find the transport layer familiar. A likely trade-off is that emotional TTS and broad dialect coverage are hard to evaluate from a specification alone, and per-minute transcription costs scale with call volume.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Mistral AI launched Voxtral 3, a new family of language models built for real-time voice agents. The system combines speech-to-text, text-to-speech, and a voice agent SDK, supports more than 50 languages, and lets agents tap into external tools during live conversations. It can run on your own infrastructure or through EU-hosted APIs to keep data under your control. The article Mistral AI launches Voxtral 3, a full voice agent stack that listens, thinks, and talks in real time appeared first on The Decoder .