Mistral launches Voxtral 3 speech-to-text and text-to-speech models and a voice agent SDK

Archive entry: this covers an event from October 20, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On October 20, 2026, Mistral AI released the Voxtral 3 family: the Small, Mini and Realtime transcription models, a Mini text-to-speech model and a voice agent SDK built on LiveKit. The release appeared in the Mistral news feed.

What Voxtral 3 includes

• Mistral described Small Transcribe 3 as state of the art in multilingual transcription, diarization, context biasing and end-of-turn detection across 50 languages, including the 24 official EU languages, and said it outperformed Whisper, Scribe v2 and Gemini 3.5 Transcribe.
• Mini Transcribe 3 had 4B parameters, and Realtime Transcribe 3 had sub-500 ms latency in 35+ languages.
• Mini TTS 3 covered 35+ languages and 50+ dialects at $13 per million characters, with voice cloning from 30+ minutes of sampled audio, Pro Voice adaptation from 30 minutes of audio and voice presets in 18 languages.
• The voice agent SDK was built on LiveKit and completed each turn in under one second. It handled SIP and WebRTC, with function calling, web search and retrieval, and offered telephony, browser and custom-audio interfaces.
• Transcription prices were $0.0035/min for Mini Transcribe 3 batch, $0.008/min for Mini Transcribe Realtime 3 and $0.0040/min for Small Transcribe 3. Voice agents cost $0.04/min with Mistral Small 4 and $0.08/min with a larger model.
• Model IDs were voxtral-small-transcribe-3-0, voxtral-mini-transcribe-3-0, voxtral-mini-transcribe-realtime-3-0 and voxtral-mini-tts-3-0, all in Public Preview.

What it means for people building with AI

Developers building voice apps could get transcription from $0.0035 per minute and a sub-second voice agent stack from one vendor. The family covered speech-to-text, text-to-speech and an agent SDK with telephony, browser and custom-audio interfaces, so the main pieces of a voice product came from the same provider.

Availability and sources

All four models were in Public Preview on the Mistral API, through European, US and global endpoints, and Mistral also offered self-deployment. More details are in the Voxtral 3 announcement and the Mistral changelog.

Sources

• Mistral Blog (Mistral AI, October 20, 2026, official)
• Voxtral 3: Speech-to-Text, TTS & Voice Agents | Mistral (Mistral AI, official)
• Changelog | Mistral Docs (Mistral AI, October 20, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.