Microsoft released MAI-Transcribe-2-Streaming on October 2, 2026, saying it enables real-time transcription for voice agents. It is the company's first major update to its transcription tools since its initial release in 2024.
Microsoft reported the model ranks first for accuracy on Artificial Analysis, measured against other models in the same category. That compares with its predecessor, which ranked third in the same benchmark.
MAI-Transcribe-2-Streaming is built on Microsoft's AI infrastructure and targets voice agents requiring instant feedback during conversations. Availability begins in October 2026, initially for developers and enterprise users.
"This model lets voice agents respond while someone is still mid-sentence," said Maximilian Schreiner, a Microsoft AI researcher. "It significantly improves the responsiveness of AI-powered voice assistants by delivering partial results in under 100 milliseconds."
The announcement follows Microsoft's recent focus on enhancing AI tools for voice interactions. Microsoft framed the update as a step toward more natural and efficient human-AI communication.
Microsoft did not say how the models will perform in complex real-world scenarios, and raised questions about the potential for misuse. The company said it will continue to refine the models and expand their capabilities.
Source: thedecoder
