MAI-Transcribe 2 is now callable through OpenRouter

Microsoft’s MAI-Transcribe 2 speech-to-text model is available through OpenRouter. In its September 3 post, OpenRouter lists support for 60 languages with automatic language identification and code switching, meaning a recording can change languages without a separate transcription workflow for each one.

The post also lists speaker diarization, which labels who is speaking, and word-level timestamps, which record when each word occurred. Together these make a transcript more useful for reviewing conversations or preparing captions. The model page shows that a default response contains only text and usage; word timestamps require setting the response format to verbose_json and appear only when the routed provider supports them.