Voxtral 3
mistral ·
Mistral has launched Voxtral 3, a new family of speech models and a voice agent SDK designed for real-time, multi-step conversations across more than 50 languages. The suite includes state-of-the-art transcription models that support diarization and context biasing, alongside text-to-speech capabilities that adapt emotion and pacing to conversational contexts. A key innovation is the integration of listening, reasoning, tool calling, and speaking into a single streaming pipeline, enabling agents to respond with sub-second latency. This architecture allows voice agents to handle complex workflows, such as booking appointments or verifying identities, while maintaining natural human-like interactions. The technology supports over 35 languages and 50 dialects, featuring voice cloning from just 30 minutes of audio for professional use cases. Organizations can access these tools via Mistral’s EU-hosted API or deploy them on their own infrastructure to ensure data sovereignty. Pricing starts at $0.0035 per minute for batch transcription and $13 per million characters for speech generation. This release aims to provide reliable, expressive, and secure voice AI solutions for contact centers and virtual assistants.
mistral