Voxtral 3 folds listening, reasoning and speaking into one stream, finishing each voice turn in under a second
Synopsis
Mistral released Voxtral 3, which puts speech recognition, reasoning, tool calling and speech synthesis into a single real-time streaming pipeline, alongside a voice agent SDK. Previously, making an agent listen accurately, reason, speak naturally and act across systems in multiple steps remained difficult. Its Small transcription model beats Whisper, Scribe v2 and Gemini 3.5 Transcribe on multilingual transcription and diarization in English, French and Spanish, and realtime transcription runs below 500 ms latency.
Interpretation
A voice agent moves from stitching separate stages together to one real-time streaming pipeline that listens, reasons, calls tools and speaks, aimed at contact centers, service desks, virtual assistants and hands-free copilots. Speech transcription and voice generation had each advanced, but making all four abilities work together reliably in real time remained a challenge. The Small transcription model reaches state-of-the-art multilingual transcription and diarization in English, French and Spanish, outperforming Whisper, Scribe v2 and Gemini 3.5 Transcribe.
The pipeline splits work across separate models: a realtime transcription model listens and detects when the caller has finished, a general model reasons and calls tools, and a speech model replies in the tone of the conversation so far. The SDK is built on LiveKit and runs speech in, reasoning and speech out as one streaming pipeline, so no stage waits for the one before it. Each turn completes in under one second and the agent stops talking when the caller cuts in; during a call it can verify identity, check a bill or book an appointment, and it connects over SIP and WebRTC to telephony, browsers and existing contact-center software.
Perspective
The report covers the Voxtral 3 release: described capabilities of transcription, synthesis and the voice agent SDK, language coverage, pricing and deployment. Intended readers are engineers and researchers building real-time voice interaction systems, in contact centers, virtual assistants and hands-free copilots. It runs through the Mistral API or self-deployment on own infrastructure, keeping sensitive conversations in the environment and controlling whether audio is retained.
The leading multilingual transcription and diarization result covers English, French and Spanish; other languages lack an equivalent comparison. Expressiveness and instruction-following claims rest on the releaser's evaluation. The language coverage of sub-500 ms realtime transcription is inconsistent in the source: one passage says over 35 languages and another says 40+, so the exact figure remains to be confirmed. Pro Voice reproduces a speaker's character from 30 minutes of audio as a professional-scenario adaptation.