Microsoft AI has launched MAI-Transcribe-2, a new automatic speech recognition model now available in public preview on Microsoft Foundry. The model supports 60 languages and introduces features such as speaker diarization, word-level timestamps, and vocabulary bias, allowing developers to pre-populate names, drug names, or industry terminology to reduce misrecognition of rare terms.
Speed and Accuracy Gains
Speed is the standout improvement. In non-real-time transcription tests by Artificial Analysis, MAI-Transcribe-2 achieved approximately 410.7x real-time speed, meaning a one-hour recording can be processed in about 9 seconds. Comparatively, GPT-Transcribe ran at 40x, and Gemini 3.5 Transcribe at 89.8x. On accuracy, the model posted an average word error rate (WER) of 5.2% on FLEURS across 60 languages, ranking first. In a separate English non-real-time test, its WER was 2.0%, placing second behind Alibaba's Fun-Realtime-ASR-preview at 1.7%.
Pricing and Availability
The introductory price is set at $0.10 per hour, a significant drop from the $0.36 per hour for the previous generation MAI-Transcribe-1.5. However, this is a limited-time offer valid only until December 31, 2026, after which the official pricing has not been announced.