Microsoft says Diagnostic Orchestrator diagnoses four times more accurately than doctors

On case reports taken from the New England Journal of Medicine, the system reached about 83 % diagnostic accuracy, while generalist physicians using references scored roughly 36 % and those without references about 22 %. In a separate set of emergency‑department admissions from Beth Israel Deaconess Medical Center, the AI achieved 80 % accuracy versus an average of 47 % for physicians, and it estimated lower costs per diagnosis.

Researchers note the tests used historical cases, not live patients, and that clinical validation is still required. Microsoft plans to begin real‑world testing at Beth Israel Deaconess in the next few months, but has not set a timeline for commercial rollout. The company frames the work as a step toward its broader “medical superintelligence” vision.