Microsoft launches real-time speech model ranked first in Artificial Analysis tests
Microsoft AI released MAI-Transcribe-2-Streaming on October 1, 2026, describing it as the company’s first streaming speech-to-text model. It is designed for voice agents, live captions and dictation, producing provisional text while a speaker is still talking and later replacing it with a stable transcript. The model supports 60 languages and continuously detects language changes automatically.
Artificial Analysis placed the model first among 38 systems in its AA-WER Streaming evaluation. It recorded a 2.5% word error rate for final transcripts at 0.13 seconds after speech ended, and the same rate for first partial transcripts at 0.12 seconds. The test used about eight hours of audio from AA-AgentTalk, VoxPopuli and Earnings22. Microsoft said internal testing showed words appearing twice as fast as with its closest competitor, while this claim was not part of the Artificial Analysis measurement.
The introductory price is $0.54 per audio hour through the end of 2026, compared with $0.10 for batch MAI-Transcribe-2. Developers can use the Realtime API or Azure Speech SDK, with access also offered through MAI Playground, Vercel and Azure Voice Live. The model is in public preview, without an SLA or open weights.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.