Mistral AI Releases Voxtral Transcribe 2 Models
Mistral AI has launched Voxtral Transcribe 2, introducing two speech-to-text models: Voxtral Mini Transcribe V2 for batch processing and Voxtral Realtime for live streaming. The Mini model offers reduced word-error rates, speaker diarization and word-level timestamps across 13 languages including Portuguese, while Realtime delivers about 200 ms latency and a 4-billion-parameter open-weights option under Apache 2.0 for on-device privacy.
Key Points
- 1Introduces two speech-to-text models — Mini Transcribe V2 (batch) and Realtime (low-latency live)
- 2Offers approximately 200 ms latency and a 4-billion-parameter Realtime model published with Apache 2.0 open-weights
- 3Enables on-device deployment, speaker diarization, and word-level timestamps for multilingual transcription workflows
Scoring Rationale
High novelty and practical on-device open-weights release, constrained slightly by company-specific product scope and no third-party benchmarks.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
