Mistral · Voice AI · October 2026

Voxtral 3: Specs, Pricing, API & Everything Announced

On October 20, 2026, Mistral AI launched Voxtral 3, a full voice agent stack built for real-time conversation. Instead of stitching separate tools together, Voxtral 3 packs listening, reasoning, tool use, and speaking into a single streaming pipeline: three transcribe models, an expressive text-to-speech model, and a voice agent SDK built on LiveKit. Here are every official spec, benchmark claim, and price from Mistral's announcement.

At a Glance

The Transcribe Lineup: Small, Mini, and Realtime

Voxtral 3 ships three speech-to-text models for different jobs.Voxtral Small Transcribe 3 is the flagship batch model, covering 50 languages — including the 24 official EU languages plus many low-resource Asian and Indic languages. It adds three production features: diarization (who spoke when), context biasing (feed it key terms and free-form text to improve recall on domain vocabulary without fine-tuning), and end-of-turn detection so agents can respond without awkward pauses. Mistral claims state-of-the-art results on multilingual transcription and diarization, outperforming Whisper, Scribe v2, and Gemini 3.5 Transcribe on the FLEURS benchmark.

Voxtral Mini Transcribe 3 is a 4B-parameter model tuned for fast, accurate batch transcription at the lowest price in the family: $0.0035 per minute. It supports the same context biasing as the Small model.

Voxtral Mini Transcribe Realtime 3 is the one built for agents: high-quality transcription at sub-500ms latency across 35+ languages, with diarization, context biasing, and native end-of-turn prediction that combines acoustic and semantic signals — designed to minimize false cut-offs when the user is only pausing mid-thought.

Voxtral Speech 3: TTS That Adapts to the Conversation

The text-to-speech side — listed in Mistral's docs as Voxtral Mini TTS 3 and priced as Voxtral Speech 3 — supports 35+ languages and 50+ dialects, with the model adjusting emotion, pacing, and emphasis based on the conversational context instead of reading flat. It is available in batch and realtime through the same API.

Mistral ships explicit instruction-following for speech: variables like emotion, pacing, and emphasis can steer delivery, alongside contextual TTS, a vocabulary library, and voice cloning from 30+ minutes of sampled audio. For professional-grade work, Pro Voice reproduces a speaker's distinctive character, emotional range, and speaking style across languages from 30 minutes of reference audio. Voice presets ship in 18 languages, from Arabic and Mandarin to Brazilian Portuguese and Vietnamese.

The headline pricing move: $13 per 1M characters. Mistral says that undercuts comparable models from OpenAI (TTS-1), Deepgram, ElevenLabs Flash, and Gemini Flash — a direct shot at the incumbents on quality per dollar.

The Voice Agent SDK: One Streaming Pipeline

The SDK — built on LiveKit — is what makes Voxtral 3 a stack rather than a model drop. The pipeline: Voxtral Mini Transcribe Realtime 3 listens and detects when the caller has finished speaking, a general model like Mistral Small reasons and calls tools, and Voxtral Speech 3 replies in the tone of the conversation so far. No stage waits for the one before it, so each turn completes in under one second — and the agent stops talking when the caller cuts in (barge-in).

During a call, agents can verify a caller's identity, check a bill, or book an appointment using function calling, web search, and retrieval over your own systems. The SDK answers and places calls over SIP, runs in browsers and apps over WebRTC, and plugs into existing telephony, IVR, and contact-center software. Because each stage is a separate model, you can swap the LLM, the voice, or the language without rewriting the agent.

Deployment is a deliberate selling point: run the whole stack on Mistral's EU-hosted API or on your own infrastructure, and control where conversations are processed and whether any audio is retained — the data-sovereignty pitch Mistral uses across its model line.

Official Pricing

Prices from Mistral's official Voxtral 3 announcement. Voxtral 3 is available today through the Mistral API, and voice agents, transcription, speech generation, and voice cloning can be tested in the Studio playgrounds.

FAQ