Skip to content

AI & Automation1 min read

Mistral AI releases Voxtral 3 for real-time voice agents

Mistral AI launched a new model family that allows businesses to build voice agents with integrated tool use.

by Fini MironFounder @ Balkaris
An abstract composition printed flat in deep teal, pale sand, muted slate blue and bright marigold orange: one large quarter circle anchored in the top left corner and a single straight bar running clear across it from the left edge to the right, the rest one flat field.
Drawn for this article on our own machine. It is not a photograph.

Mistral AI launched Voxtral 3, a family of language models designed for real-time voice agents, according to The-decoder. The system includes a speech-to-text model, a text-to-speech model, and a voice agent SDK based on LiveKit. It combines listening, thinking, tool use, and speaking into a single streaming pipeline. The transcription model covers more than 50 languages, while the text-to-speech output supports over 35 languages and 50 dialects. Mistral AI states the system can run on private infrastructure or through EU-hosted APIs.

Why it matters

Swiss businesses can now deploy voice agents that interact with internal tools like booking systems or invoice databases. The ability to host these models on private infrastructure addresses specific data residency requirements. You can choose between local control or EU-hosted APIs to manage where your information stays.

  • Reduced latency for customer service automation.
  • Support for multi-dialect communications in a single pipeline.
  • Lower entry costs for transcription starting at $0.0035 per minute.

What to watch

Check the latency of the streaming pipeline for your specific use case. Real-time interaction depends on how quickly the model processes tool use during a live call. You should also verify if the emotional adaptation in the text-to-speech output aligns with your brand voice before moving to production.

Swiss firms will move voice automation to private servers by 2026. This happens if local data laws remain strict and low-latency performance stays consistent.

Key takeaways

  • The system supports over 50 languages for transcription and 35 languages for speech output.
  • Businesses can host the models on private infrastructure to meet data residency requirements.
  • Transcription costs start at $0.0035 per minute.

Frequently asked

How does the system handle specific business tasks like checking an invoice?

The models use a tool-use capability to interact with internal databases or booking systems during a live call.

Will the voice output sound natural for my specific brand?

You must verify if the emotional adaptation in the text-to-speech output aligns with your brand voice before production.