Microsoft unveils AI Diagnostic Orchestrator that outperforms doctors in tests

Microsoft's Diagnostic Orchestrator, an AI platform that simulates a panel of medical experts, was tested against human clinicians on thousands of real-world cases. Powered by GPT-5 and a suite of cooperating agents, it achieved roughly 83% accuracy on cases drawn from The New England Journal of Medicine and 80% on emergency department records from Beth Israel Deaconess Medical Center, outperforming generalist doctors who scored 36% and 47% respectively.

The system also proposed lower-cost diagnostic strategies. Microsoft compiled the test set into an open-source benchmark called SDBench and intends to begin real-world hospital trials at Beth Israel Deaconess within months. While the company acknowledges the need for clinical validation before widespread deployment, it views the results as a major milestone toward its goal of "medical superintelligence."