Mistral AI has released Voxtral 3, a family of models built for real-time voice agents. It includes a speech-to-text model, Voxtral Small Transcribe 3, and a text-to-speech model, Voxtral Mini TTS 3, along with a voice agent SDK based on LiveKit. The parts are combined into a single streaming pipeline that handles listening, reasoning, tool use and speaking.
Transcription covers more than 50 languages. Text-to-speech output supports over 35 languages and 50 dialects, with emotional adaptation. During live conversations, agents can reach external systems, for example to book appointments or check invoices.
Mistral says the system can run on a customer's own infrastructure or through EU-hosted APIs, which keeps data under the customer's control. Pricing for transcription starts at $0.0035 per minute. More details are available in Mistral's documentation.
Why it matters if you build agents
Developers building voice agents can get transcription, speech output and tool calls from one stack, with the option to self-host. Teams with data residency requirements may find the EU-hosted API or on-premises deployment relevant to their setup.
Check the per-minute transcription price and the supported languages against your use case before committing to the stack.
Get stories like this every weekday
One short email with the day's agent tooling news.