Voxtral 3

Released October 20, 2026 by Mistral AI, Voxtral 3 is a full voice-AI platform: three transcription models (Voxtral Mini Transcribe 3, Voxtral Small Transcribe 3, Voxtral Mini Transcribe Realtime 3), a text-to-speech model (Voxtral Speech 3), and a voice agent SDK that combines listening, reasoning, tool calling and speaking into one real-time streaming pipeline. Mistral claims the flagship transcription model posts a 4.1% word error rate on the FLEURS 45 multilingual benchmark — clearing Whisper large-v3, Scribe v2 and GPT Transcribe on multilingual accuracy — with sub-one-second turn latency and natural interruption handling.

Key facts: the models run through the Mistral API on European, US and global endpoints, and can also be self-deployed on private infrastructure. Transcription supports 50 languages (including all 24 official EU languages) with speaker diarization and context biasing for domain-specific terms; the realtime tier advertises sub-500ms latency. Voxtral Speech 3 covers 35+ languages and 50+ dialects, adjusts emotion, pacing and emphasis to the conversation, and its Pro Voice feature clones a voice and style across languages from 30 minutes of audio. The SDK is built on LiveKit: phone calls connect over SIP, browser and app sessions over WebRTC, and each pipeline stage is a separate model, so the LLM, voice or language can be swapped without rewriting the agent.

Plans: pay-as-you-go API pricing — Voxtral Mini Transcribe 3 batch at $0.0035/min, Small Transcribe 3 at $0.004/min, Mini Transcribe Realtime 3 at $0.008/min; Voxtral Speech 3 at $13 per 1 million characters. The full voice agent API costs $0.04/min with Mistral Small 4 handling reasoning, or $0.08/min with another reasoning model. This is a developer API product — there is no consumer free app.

Best for: developers building phone and in-app voice agents, call-center and contact-center automation, and multilingual transcription pipelines. European teams get EU-hosted endpoints for sovereignty requirements. Note: all accuracy and cost-comparison claims are Mistral's own — wait for independent benchmarks before committing. Teams deploying voice agents in the EU should also check EU AI Act transparency and audio-watermarking obligations.

Related: ElevenLabs — Eleven v4 · Suno Speech · ChatGPT

Pros

• A full voice-AI platform: three transcription models, an expressive text-to-speech model, and a voice-agent SDK in one real-time streaming pipeline.
• Transcription supports 50 languages with speaker diarization and context biasing for domain-specific terms.
• Speech covers 35+ languages and 50+ dialects, adjusting emotion, pacing and emphasis, with Pro Voice cloning across languages from 30 minutes of audio.
• Models run through the Mistral API on European, US and global endpoints, and can also be self-deployed on private infrastructure.

Watch out for

• All accuracy and cost-comparison claims are Mistral's own — wait for independent benchmarks before committing.
• This is a developer API product; there is no consumer free app.
• Teams deploying voice agents in the EU should check EU AI Act transparency and audio-watermarking obligations.

Best for

Developers building phone and in-app voice agents, call-center and contact-center automation, and multilingual transcription pipelines. European teams get EU-hosted endpoints for sovereignty requirements. Who should skip it: non-developers looking for a ready-made consumer voice app.

Pricing

Pay-as-you-go API pricing — Voxtral Mini Transcribe 3 batch at $0.0035/min, Small Transcribe 3 at $0.004/min, Mini Transcribe Realtime 3 at $0.008/min; Voxtral Speech 3 at $13 per 1 million characters. The full voice agent API costs $0.04/min with Mistral Small 4 handling reasoning, or $0.08/min with another reasoning model — check the official pricing page for current pricing.

FAQs

What is Voxtral 3? Released October 20, 2026 by Mistral AI, it is a voice-AI platform covering transcription, expressive speech generation and a full voice-agent SDK.

How many languages does it support? Transcription supports 50 languages, and speech covers 35+ languages and 50+ dialects, according to this page.

How do voice agents connect? The SDK is built on LiveKit: phone calls connect over SIP, and browser and app sessions over WebRTC.

Is there a free consumer app? No. This is a developer API product — there is no consumer free app.