Mistral has introduced Voxtral 3, a new family of speech-to-text and text-to-speech models paired with a voice agent SDK. The system brings listening, reasoning, tool calling, and speaking into a single real-time streaming pipeline, designed for natural conversational experiences across contact centers, service desks, virtual assistants, and hands-free copilots.
The speech-to-text model, Voxtral Small Transcribe 3, is described as state-of-the-art in multilingual transcription. It supports 50+ languages, identifies who said what (diarization), uses keyphrases to improve accuracy (context-biasing), and detects when a speaker has finished so agents can respond without delay. The expanded language list includes 24 official EU languages and many low-resource Asian and Indic languages.
The text-to-speech model, Voxtral Mini TTS 3, offers new voices that adjust emotion, pace, and emphasis based on conversational context, with support for 35+ languages and 50+ dialects. For professional use cases, Pro Voice adaptation can reproduce a speaker's distinctive character, emotional range, and speaking style across languages using just 30 minutes of audio.
Voxtral 3 is designed for sovereign deployment. It is available through Mistral's EU-hosted API or for self-deployment on an organization's own infrastructure, giving control over where audio is processed, how data is handled, and whether conversations are retained. This makes it suitable for sensitive conversations that must stay within a controlled environment.
Source: Hacker News · Summarized by HeadlinesBriefing