Model releases & benchmarks
Our first streaming transcription model debuts at no. 1 on Artificial Analysis
Microsoft introduced MAI-Transcribe-2-Streaming alongside multilingual MAI-Voice-2.1 and its faster Flash variant.
Why it matters: For agent builders, streaming partial transcripts could let an agent begin reasoning or calling tools before a speaker finishes, while pairing transcription with Flash is intended to reduce delays at both ends of the voice interaction. Microsoft presents these models as building blocks for conversational agents; the performance comparisons and evaluations cited are the company’s claims.