A full stack for voice agents
Mistral AI has announced Voxtral 3, a family of language models designed for building real-time voice agents. The system combines speech-to-text, text-to-speech and a voice agent SDK in a single stack.
What it offers
- Support for more than 50 languages
- Access to external tools during live conversations
- The ability to run on your own infrastructure
- EU-hosted APIs to keep data under your control
Why it matters
Voice agents are bringing AI into areas from customer service to call centers. Voxtral 3's combination of listening, thinking and talking in real time aims to make these applications feel more natural and seamless. Mistral's emphasis on European infrastructure also offers an alternative for organizations that care about data sovereignty.
