A clearer view of AI.
THE INDEPENDENT AI.

WORTH TESTING · 21 OCTOBER 2026

Mistral's Voxtral 3 launch demo: a voice agent that copes with real conversation

Watched in full (2 minutes). It's Mistral's official launch demo: two people talk to a voice agent called Charlotte, which tells them apart, gets interrupted without losing track, and switches between English, French and German in one reply with a change of tone. Mistral explains it as three stages: listen, think (with Mistral Large) and speak. The sub-second speed can't be checked in an edited video, and the web search it mentions isn't actually shown. Verdict: a launch demo rather than a tutorial, but a good benchmark for what any voice agent should handle. No coding needed to talk to one; building one needs intermediate coding skills or a developer. It has nine real things to test.

Try this

  1. Put two people on one call and check the agent tells them apart. 00:06
  2. Think of it as three swappable stages: transcribe, reason, speak. 00:30
  3. Interrupt the agent mid-sentence and see whether it stops cleanly. 00:44
  4. Check it still remembers the conversation after being interrupted. 00:52
  5. Give it an instruction with several steps in one go. 01:00
  6. Ask for a reply that switches between languages, such as English, French and German. 01:00
  7. Ask it to change its tone, for example from matter-of-fact to mildly amused. 01:00
  8. Have a developer try tool calling through the SDK, so it can look things up mid-call. 01:21
  9. Time the reply delay yourself against Mistral's under-a-second target. 01:44

Timestamps point to the moment in the source video.

Cost

Free to watch on X. Our notes don't cover the cost of the tools shown.

Source

Related AI tools

How we knowSteps to try, not a claim that we independently tested every result. Follow the linked sources and check current prices. Our methods & sources ↗