Microsoft’s AI diagnostic system got complex medical cases right more than twice as often as generalist physicians in a simulated setting. That finding comes from a peer-reviewed study published in the journal Science.
A virtual panel of doctors, minus the scheduling conflicts
The system is called MAI-DxO, short for Microsoft AI Diagnostic Orchestrator. It coordinates several AI agents that behave like a team of physicians gathered around a hard case. The agents question the patient, decide which tests to order, and argue among themselves before settling on a diagnosis. The process is iterative, so each new answer or test result shapes the next step. The system also weighs cost as it reasons, considering whether a test is worth ordering, not just whether it might be useful.
On complex cases drawn from the New England Journal of Medicine, MAI-DxO reached accuracy of approximately 83%. Generalist physicians scored between 22% and 36%, with the range depending on how much access they had to medical references.
Across more than 300 emergency department cases, the system hit 80% accuracy, while generalist physicians who had web access reached 47%.
In some configurations, estimated diagnostic costs came in up to 20% lower than those of the physicians.
From a 2025 preprint to a Science paper
Microsoft first showed off the system in 2025 research, which reported an 85.5% performance figure on NEJM cases. The new study validates and extends that earlier work. The headline accuracy number on NEJM cases is slightly lower this time, at approximately 83%, but the evidence base is broader, and it now includes the emergency department cases.
The effort is led by Mustafa Suleyman, CEO of Microsoft AI, who has framed the results as a significant step toward what he calls “medical superintelligence.”
MAI-DxO is model-agnostic, meaning the orchestration layer is designed to work with different underlying AI models rather than being tied to one.
Microsoft also plans to open-source SDBench, the benchmark used to test the system, which would let academic researchers and others run their own evaluations.
What this means for hospitals, doctors, and Microsoft
Microsoft says real-world hospital testing is set to begin in the coming months. The study pitted the AI against generalist physicians, not the specialists who typically handle the rarest and hardest cases.
Microsoft has pitched the system as a way to ease the diagnostic drudgery clinicians deal with, framing it as support for doctors rather than a replacement for them. The company has also signaled aspirations to bring the technology into consumer products like Copilot.
Microsoft says clinical validation and regulatory compliance will come before any wider rollout.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.