At Ignite 2026, Microsoft announced that its medical AI product Dragon Copilot Transcribe now runs entirely on its in-house MAI-Transcribe-2 speech recognition model, cutting multilingual workflow error rates by roughly 50% across about 10 million patient visits per month and reaching roughly one-third of U.S. physicians. The model posts a 5.2% average word error rate across 60 languages on the FLEURS benchmark, supports speaker diarization, word-level timestamps and code-switching, and processes long audio up to 10x faster than some competitors.
Modelsslug: mai-transcribe-2
MAI Transcribe 2
Microsoft's in-house multilingual speech transcription model with diarization, word-level timestamps and code-switching support.
Curated Intelligence
IT之家 · 智能时代T2·Models
74 pts
OpenRouter (@OpenRouter)T1·Tools
58 pts
OpenRouter reports MAI Transcribe 2 weekly requests climbed from 0.7M in its first week to 8.5M, a roughly 12x increase. The broader speech-to-text category grew 3x over the same period, signaling category-wide acceleration.