What changed
Microsoft released MAI-Transcribe-2 for batch speech recognition with speaker diarization, word-level timestamps, keyword biasing, configurable verbatim or clean styles, code switching and automatic language detection across 60 languages. The launch price was $0.10 per audio hour for a limited period.
Why it matters
A low-cost transcription model with diarization and formatting controls can reduce pipeline complexity for meeting, media and support workflows. Accuracy, consent, speaker attribution and domain vocabulary still need workload-specific tests before regulated use.
What to do
Evaluate a consented, representative audio set with noise, accents, code switching and overlapping speakers. Measure word error, speaker attribution, timestamps, omissions and cost; keep human review for clinical, legal, accessibility or compliance outputs.
Caveat
The 5.2% FLEURS word-error rate and comparative speed claims come from Microsoft and Artificial Analysis configurations, not this catalog's independent replication. The $0.10 hourly price was described as limited-time, and availability and pricing can differ by provider.