Mistral AI launches Voxtral 3, a full voice agent stack that listens, thinks, and talks in real time
decoder ·
Mistral AI has introduced Voxtral 3, a comprehensive voice agent stack designed for real-time interaction. This new model family integrates speech-to-text transcription, text-to-speech generation, and a voice agent SDK built on LiveKit into a single streaming pipeline. The system supports transcription in over 50 languages and offers text-to-speech capabilities in more than 35 languages with 50 dialects, including emotional adaptation features. Agents are equipped to utilize external tools during live conversations, such as booking appointments or checking invoices. Mistral emphasizes data privacy by allowing the system to run on private infrastructure or through EU-hosted APIs. Pricing for the transcription service starts at $0.0035 per minute. This launch positions Voxtral 3 as a complete solution for developers seeking to build responsive voice-based AI applications.
decoder