Voxtral 3🇫🇷
Mistral AI
Voice assistants have to listen, think and talk quickly enough that a conversation doesn't feel stilted. Voxtral 3 is Mistral's set of speech models for that job: transcribing audio, following speech as it happens, and reading text aloud. Alongside it comes a toolkit for developers building voice agents that can be interrupted mid-sentence and can take actions while they talk.
What's new
- Four models: two that transcribe recordings, one that transcribes live speech and one that turns text into speech
- Transcription covers more than 50 languages, with speaker labels and the option to favour your own vocabulary
- Text-to-speech covers more than 35 languages and supports voice cloning
- The voice agent toolkit handles interruptions, function calls, web search, phone calls and browser calls
- Available through Mistral's API on European, US and global endpoints, priced per minute of audio
Why it matters
Good speech tools are a big part of useful voice assistants, and many developers building them today rely on non-European services. This gives European teams another option, including European API endpoints. Claims about speed, accuracy and voice quality come from Mistral, so independent comparisons are still to come.
About the company
Mistral AI is a pioneering French artificial intelligence startup established in April 2023 by researchers with backgrounds in leading AI organizations. The company aims to democratize access to advanced AI models and solutions, enabling a broad spectrum of users to leverage AI for various applications.