Microsoft’s most powerful transcription AI model: MAI-Transcribe-2 is in use, with an average word error rate of 5.2% for 60 languages.
Microsoft released MAI-Transcribe-2, claiming it to be the fastest, most accurate, and cheapest AI speech-to-text model, with a limited-time price of $0.10 per hour. This model supports 60 languages and natural mixed conversations. In the FLEURS test set, its average word error rate was 5.2%, lower than Gemini 3.1 Pro (5.3%), GPT-Transcribe (10.4%), and Whisper v3-Large (22.8%). In terms of performance, it is 10 times faster than GPT-Transcribe, 7 times faster than Scribe v2, and 5 times faster than Gemini 3.5 Transcribe. New features include speaker separation, word-by-word timestamps, and configurable transcription styles (verbatim and clean modes), as well as support for keyword offset, automatic language recognition, and transcription in noisy environments. Developers can use Microsoft Foundry, MAI Pl…