Mistral Releases Voxtral Transcribe 2 Models

Mistral recently released Voxtral Transcribe 2, a family of two speech-to-text models including the open-weights Voxtral-Mini-4B-Realtime-2602 (Apache-2.0, 8.87GB on Hugging Face) and a closed-weight voxtral-mini-latest accessible via the Mistral API. The company updated its API console with a speech-to-text playground offering diarization and downloads in text, SRT, and JSON; the release follows the original Voxtral from July 2025.