French AI firm Mistral AI launches two new speech-to-text models, Voxtral Mini Transcribe V2 for batch processing and Voxtral Realtime for low-latency streaming. Both offer high accuracy, speaker diarization, and support 13 languages. The real-time model features a delay as low as 200ms and can run locally on devices for enhanced privacy. Pricing starts at $0.003 per minute for batch transcription.