Microsoft AI Slashes Transcription Cost to 10¢/Hour

Microsoft AI put a number on the table Thursday that's hard to ignore: 10 cents per hour of audio for MAI-Transcribe-2, its newest speech-recognition model. Five months ago, the company's first model in this line cost $0.36 an hour. The new price cuts that by roughly 72%, and Microsoft says the model is also faster and more accurate than what OpenAI, Google, and ElevenLabs currently offer. For a bank or telecom running 100,000 hours of call-center audio a year, that shift moves the annual bill from $36,000 to $10,000, the kind of drop that changes how a finance department thinks about transcription altogether.

The release lands inside a broader pattern at Microsoft: build frontier models modality by modality, then swap them into products that once leaned on OpenAI's technology. Transcription has moved faster than any other piece of that plan, and MAI-Transcribe-2 is the sharpest evidence of it so far. It also signals something about how Microsoft, having spent $13 billion cultivating its OpenAI partnership, now wants to compete in AI on its own terms. The model's expanded language support and its rollout across Microsoft's developer tools point to where that ambition is headed next.