Mistral releases Voxtral 3 with a voice agent SDK
Transcription in 50 languages and speech in more than 35, plus a toolkit for building voice agents.
Mistral has released Voxtral 3, a family of speech-to-text and text-to-speech models, together with a voice agent SDK. The SDK, built on LiveKit, joins listening, reasoning, tool calling and speaking into one real-time streaming pipeline. Mistral says the transcription model covers 50 languages and can tell speakers apart, while the speech model supports more than 35 languages and adjusts emotion, pace and emphasis to the conversation.
For teams building phone or web assistants, one stack can hear a caller, look things up in their own systems and answer, with each turn completing in under a second according to Mistral. The models run through Mistral's EU-hosted API or on your own infrastructure, so you decide where audio is processed and whether conversations are kept.
Voxtral 3 is available now through the Mistral API, with European, US and global endpoints. Mistral lists batch transcription from $0.0035 per minute and speech generation at $13 per million characters. If you handle sensitive calls, compare the hosted and self-deployed options before choosing.