Microsoft Releases Real-Time Transcription AI MAI-Transcribe-2-Streaming
Microsoft is shipping a streaming transcription model into a crowded field and pricing it by the hour, with the accuracy and latency ranking coming from a third-party analysis rather than its own testing.
Reporting from 1 source: GIGAZINE.
Microsoft released MAI-Transcribe-2-Streaming on October 1, 2026, its first real-time transcription model. Analysis by the third-party organization Artificial Analysis found it has the lowest error rate among the real-time transcription models tested. It has shorter latency than models of equivalent accuracy. The model is available through an API at $0.54 per hour until the end of 2026.
Microsoft put a price on the accuracy claim: $0.54 per hour through the API until the end of 2026. The model is its first built for real-time transcription, not batch processing.
The ranking behind the release comes from Artificial Analysis, which placed MAI-Transcribe-2-Streaming at the lowest error rate among the real-time transcription models it tested. The same analysis found shorter latency than models of comparable accuracy. Both numbers come from that third-party evaluation, not from Microsoft's own benchmarks.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.