Mistral AI Releases Voxtral 3 Voice Suite With Cloning And Real-Time Transcription
A single API now covers the whole audio chain from live transcription to cloning a named speaker from half an hour of reference audio, with Mistral pricing real-time transcription at $0.008 per minute.
Reporting from 1 source: GIGAZINE.
Mistral AI has released Voxtral 3, a suite of four voice models: Voxtral Small Transcribe 3 and Voxtral Mini Transcribe 3 for transcription in 50+ languages, Voxtral Mini Transcribe Realtime 3 for real-time work at 240 milliseconds of latency, and Voxtral Speech 3 for synthesis in 35+ languages and 50+ dialects. Thirty minutes of reference audio can clone a specific person's voice.
The four models sit under one name and one API. Small carries the highest transcription accuracy of the Transcribe models, a naming choice Mistral notes is confusing next to Mini, and Mistral's comparison graph puts both above the prior Voxtral Mini Transcribe 2 on error rate. The real-time model transcribes more accurately than GPT Live Transcribe at 240 milliseconds of latency, with accuracy improving as latency rises.
Speech 3 reads proper nouns and numbers at a cost Mistral compares to Gemini 3.8 Flash TTS. Cloning needs 30 minutes of reference audio. Pricing runs from $0.0035 to $0.008 per minute for the transcription models, and $13 per million characters for synthesis.
- Voxtral Small Transcribe 3: High-performance transcription, 50+ languages, $0.0040 per minute
- Voxtral Mini Transcribe 3: Cost-efficient transcription, $0.0035 per minute
- Voxtral Mini Transcribe Realtime 3: Real-time transcription at 240 ms latency, $0.008 per minute
- Voxtral Speech 3: Speech synthesis, 35+ languages and 50+ dialects, $13 per million characters
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.