Mistral AI Launches Voxtral 3 Voice Models and Agent SDK

Mistral AI launches Voxtral 3, a family of speech models and a voice agent SDK, as detailed in a company blog post. The system combines transcription, reasoning, tool use and speech generation in a streaming pipeline for phone and web conversations.

Voxtral Small Transcribe 3 supports more than 50 languages, speaker identification, context biasing and detection of when a speaker has finished talking. Mistral says the Small model beats Whisper, Scribe v2 and Gemini 3.5 Transcribe on its benchmarks. The 4 billion parameter Mini variant costs $0.0035 per minute, while the real time transcription model supports more than 35 languages with latency below 500 milliseconds.

Voxtral Mini TTS 3 generates speech in more than 35 languages and 50 dialects, with controls for emotion, pace and emphasis. It costs $13 per 1 million characters, and a Pro Voice option reproduces a speaker's style from 30 minutes of audio.

Voxtral 3 is available through Mistral AI's EU hosted API or for deployment on an organization's own infrastructure. The voice agent SDK can connect with business systems to retrieve information and complete tasks during live conversations.