Voxtral 3

Not yet independently verified. Rests on Mistral's own announcement, which its page dates 22 October 2026 and its news index lists under 21 October; no independent test was found, and the licence and any open-weights release are not stated. We will update this when it can be confirmed, and remove this note.

Voxtral 3 is a family of speech models from the Paris company Mistral AI, released in October 2026 together with a voice agent SDK. It covers speech-to-text, text-to-speech and a pipeline that joins them to a reasoning model so that a voice agent can listen, call tools and answer in one stream.

What it does

The family has three transcription models and one voice model. Voxtral Small Transcribe 3 is the most accurate, with transcription in more than 50 languages, speaker diarization and context biasing from supplied keyphrases. Voxtral Mini Transcribe 3 is the low-cost batch model, and Voxtral Mini Transcribe Realtime 3 streams text with end-of-turn detection, at under 500 milliseconds of latency across more than 40 languages, Mistral says. Voxtral Mini TTS 3 generates speech in more than 35 languages and 50 dialects, adjusting emotion, pace and emphasis to the conversation, and a Pro Voice option adapts a voice from about 30 minutes of a speaker's audio.

The SDK runs speech in, reasoning and speech out as one streaming pipeline, with a general model such as Mistral Small handling the reasoning and tool calls. Mistral says each turn completes in under a second and the agent stops talking when the caller interrupts. It connects over SIP for phone calls and WebRTC for browsers and apps.

Pricing

Model or service	Price
Voxtral Mini Transcribe 3 (batch)	$0.0035 per minute
Voxtral Small Transcribe 3	$0.0040 per minute
Voxtral Mini Transcribe Realtime 3	$0.008 per minute
Voxtral Mini TTS 3	$13 per 1M characters
Voice agent API with Mistral Small 4	$0.04 per minute
Voice agent API with a larger model	$0.08 per minute

Who it is for

Mistral aims Voxtral 3 at contact-centre agents, service desks, virtual assistants and hands-free copilots, and at organisations that need to control where audio is processed. It is available through Mistral's API on European, US and global endpoints, and the whole stack can be run on-premises or in a private cloud by arrangement with Mistral.

Key facts

Maker: Mistral AI Released: 22 October 2026 (Mistral's page) Models: Small Transcribe 3, Mini Transcribe 3, Mini Transcribe Realtime 3, Mini TTS 3 Languages: 50+ for transcription, 35+ for speech Deployment: Mistral API (EU, US, global) or self-deployment Licence: not stated