Voxtral Small Transcribe 3

Release Summary

Mistral's flagship Voxtral 3 transcription model, in public preview as voxtral-small-transcribe-3-0. Audio in, text out, with the model card's 32k context and no separate output cap. Batch price is $0.004 per minute. It covers 50+ languages, with diarization, context biasing, and end-of-turn detection, and the card also lists translation, speech question answering, summarization, and function calling. On the FLEURS 45 chart, macro WER is 4.1% when the language is specified and 4.0% when it is inferred, ahead of Gemini 3.5 Transcribe at 6.1% and 4.5% and level with MAI Transcribe 2 at 4.1% and 4.0%. Diarization cpWER is 18.4%. Earnings-22 keyphrase recall is 94.6% with context and 81.4% without. The cheaper file model is Voxtral Mini Transcribe 3.

Timeline

Other Voxtral versions