Voxtral Mini Transcribe 2507

Voxtral Mini Transcribe 2507 (July 2025) is the transcription-optimized configuration of the 3B Voxtral Mini, built on the Ministral-3B backbone and released under Apache 2.0. It runs in a dedicated speech-to-text mode that maximizes transcription accuracy while keeping the model small and cost-efficient. The model is natively multilingual across widely used languages, with automatic language detection, and processes audio up to 30 minutes per request. Because it shares the Voxtral Mini foundation, it retains the language understanding of its backbone while specializing in clean, accurate transcription. Voxtral Mini Transcribe is suited to meeting notes, interview transcription, and media workflows where reliable, affordable speech-to-text is the priority, and it can be deployed on modest hardware thanks to its compact footprint.