MAI-Transcribe-2-Streaming

Not yet independently verified. Every detail here comes from Microsoft AI’s own announcement of 1 October 2026. The accuracy ranking and speed comparisons are Microsoft’s statements; we have not read the ranking site’s own page, and no independent press report had been found at 2026-10-04. We will update this when it can be confirmed, and remove this note.

MAI-Transcribe-2-Streaming is a real-time speech recognition model released on 1 October 2026 by Microsoft AI, the company’s in-house model lab. It transcribes 60 languages and detects the spoken language continuously, so a conversation can switch language without reconfiguration. Instead of waiting for a speaker to finish, it returns first hypotheses, called partials, in just over 100 milliseconds, revises them as more audio arrives and then commits a stable transcript. Microsoft says this lets voice agents start reasoning or calling tools before the caller finishes speaking. Microsoft states that it ranks first for accuracy on both final and partial transcripts on Artificial Analysis, and that in its own evaluation words appear twice as fast as with its closest competitor; those are Microsoft’s claims. The introductory price is $0.54 per hour of audio through the end of 2026. It is available through Microsoft Foundry, the MAI Playground, Vercel and Azure Voice Live, with LiveKit listed as coming soon. It is a separate product from the non-streaming MAI-Transcribe-2. The weights are not published.