Mistral shipped a voice agent SDK with speech models that call tools mid-conversation across 50+ languages. It can run on an EU-hosted API or self-deployed, which matters for teams with data sovereignty constraints.
Key takeaways · AI-distilled
Mistral's Voxtral 3 voice AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → SDK is built on LiveKit and streams transcription, an LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → such as Mistral Small, and speech output without stages waiting on each other, so Mistral says each turn completes in under one second and the agent stops talking when the caller interrupts.
Mistral prices Voxtral Mini Transcribe 3 at $0.0035 per minute batch, the Realtime variant at $0.008 per minute, Small Transcribe 3 at $0.004 per minute, and Voxtral Speech 3 at $13 per 1M characters, which Mistral says undercuts OpenAI TTS-1, ElevenLabs Flash and others.
Voxtral Transcribe 3 supports context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → biasing, where key terms or free text are supplied to improve recall on domain vocabulary, which Mistral says avoids the need to fine-tuningTaking a trained model and training it a bit more on your own examples so it gets better at one specific job.Full definition → for specialized production settings.
Because each Voxtral 3 agent stage is a separate model, Mistral says teams can swap the LLM, voice or language without rewriting the agent, and calls can run over SIP telephony or WebRTC in browsers and apps.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
