4 September 2026

Microsoft releases MAI-Transcribe-2 speech recognition model

• Microsoft launched MAI-Transcribe-2, a speech recognition system that identifies who is speaking, adjusts transcription style, and marks exactly when each word appears.
• The model handles audio in 60 languages and costs 10 cents per hour of transcription, undercutting costs of existing speech recognition services.
• Microsoft is replacing OpenAI technology in its products with its own speech models as part of a broader shift toward using internally-built AI.

How it was covered

TLDR AITLDR editorial team

Microsoft released MAI-Transcribe-2, a speech recognition model with diarization, configurable transcription styles, and word-level timestamps that beats competitors. Priced at 10 cents per hour and transcribes audio in 60 languages, representing Microsoft's strategy to swap its own frontier-class models into products that once ran on OpenAI technology.