Mistral Just Launched Voxtral 3 — a Voice-Agent Stack That Talks in 50+ Languages

 

Mistral AI just dropped Voxtral 3 — not another transcription model, but a full voice-agent stack. Announced October 20, it bundles new speech-to-text and text-to-speech models with a real-time voice agent SDK, so one streaming pipeline can listen, think, use tools, and talk back. Transcription covers 50+ languages; speech output covers 35+.

What's actually new

Until now, building a voice agent meant stitching three separate pieces: one model to transcribe, an LLM to reason, another to speak. Voxtral 3 ships all three plus a LiveKit-based voice agent SDK in one connected pipeline — and because each stage is a separate model, developers can swap the LLM, the voice, or the language without rebuilding the agent. In Mistral's demo, an English-speaking caller holding a French referral letter got a clinic appointment booked and an English confirmation text promised, all mid-call.

The model lineup

• Voxtral Small Transcribe 3 — speech-to-text in 50+ languages, including all 24 official EU languages plus Asian and Indic languages. Adds speaker diarization, domain-vocabulary hints, and end-of-turn detection so agents know when to jump in.
• Voxtral Mini Transcribe 3 — a lean 4-billion-parameter transcriber for batch jobs.
• Voxtral Mini Transcribe Realtime 3 — built for live agents, with what Mistral says is sub-500-millisecond latency.
• Voxtral Speech 3 — text-to-speech in 35+ languages and 50+ dialects, adjusting emotion, pacing, and emphasis to the conversation. A Pro Voice feature adapts a speaker's voice across languages from 30 minutes of audio.
• Voice agent SDK — handles phone calls over SIP and browser/app sessions over WebRTC, and plugs into existing telephony, IVR, and contact-center software.

Pricing and availability

Batch transcription starts at $0.0035 per minute (Mini), $0.0040 for Small, and $0.008 for the realtime model. Speech generation costs $13 per million characters. The full voice-agent service runs $0.04 per minute with Mistral Small 4 doing the reasoning, or $0.08 per minute with another reasoning model. It's available today through Mistral's API on EU, US, and global endpoints — or self-hosted on your own infrastructure. It's the same open playbook Mistral used for its open-weight flagship: keep the data under the customer's control instead of shipping call audio to a US cloud.

Verdict

Honesty first: the benchmark claims — beating Whisper large-v3 on FLEURS, "state-of-the-art" diarization — are Mistral's own, with no independent verification yet. But the product shape is what matters. ElevenLabs just proved voice agents are having their moment, and Mistral's answer is to own the whole stack rather than rent pieces of it. As AI agents move from text chat to phone calls, the EU-hosted, self-deployable angle gives European enterprises a real reason to pick up the phone. Worth watching.