Microsoft claims its multi-agent AI diagnostic tool beats doctors by up to four times

Microsoft says it has built an AI system that diagnoses medical conditions up to four times more accurately than human doctors, according to The Verge. The findings were published in the journal Science on Thursday, and Microsoft AI CEO Mustafa Suleyman told The Verge that the moment healthcare AI "is finally here" — while acknowledging the system still has to prove itself beyond historical test cases.

A panel of AI agents instead of one model

The tool, called the Diagnostic Orchestrator, is built to behave like a panel of medical experts. Multiple AI agents reason through a patient's symptoms and medical history, weigh possible explanations, decide which tests to order, and settle on a diagnosis — a sequence meant to mirror how doctors actually investigate a case. According to The Verge, the system runs on OpenAI's GPT-5, and Microsoft says the coordinated agents outperformed frontier AI models operating on their own.

To test it, Microsoft assembled scenarios from published medical case reports and emergency department data into a benchmark called SDBench, which the company plans to open-source.

The numbers

On cases drawn from hundreds of reports in The New England Journal of Medicine, the orchestrator reached roughly 83 percent diagnostic accuracy. Generalist physicians scored about 36 percent when allowed to use references, and 22 percent when working without them.

The system also led on a second test set built from a large database of patient admissions at Beth Israel Deaconess Medical Center in Boston, scoring 80 percent against an average of 47 percent for human doctors. It tended to arrive at its diagnoses at a lower estimated cost than its human counterparts.

There are caveats inside the research itself. The peer-reviewed paper does not include newer models such as GPT-6 Astra, Meta's Muse Wonder Full, Google's Gemini 4 Argon, or Anthropic's Opus 5.5. The project began back when GPT-4.1 was the most advanced model available, and the researchers say subsequent testing suggests that improvements in underlying model capability do not necessarily erode the benefit of having multiple agents work together.

Tested on records, not patients

Microsoft is explicit about the gap that remains. The system has only been evaluated on past cases, and it is unclear how it would perform with live patients, incomplete information, and the pressures of a real clinical environment. "It needs clinical validation," Suleyman told The Verge. "That is a leap, but this is a major milestone along that journey."

Hospital trials and a broader agenda

Dr. Dominic King, Microsoft's AI health lead and a coauthor of the study, told The Verge that the company plans to begin testing the system in hospital settings with Beth Israel Deaconess. He said the work will likely "get underway in the next couple of months," though neither he nor Suleyman would commit to a firm timeline. Commercialization could follow, but Microsoft has not said when or how the technology would be made available, and King stressed the regulatory and compliance requirements that medicine demands.

Diagnosis is one strand of a broader program Microsoft calls "medical superintelligence" — AI that can outperform the best human-led healthcare systems. In a blog post and a separate article in BMJ Digital Health & AI, the company lays out five other areas it believes AI will need to master, including coordinating clinical care, making treatment decisions, and handling the human side of medicine.

Suleyman predicted "extremely impressive deployments of intelligent systems in production" next year, and said he expects performance indistinguishable from humans on medical superintelligence within a couple of years. At the same time, he insisted clinicians will keep a central role in patient care, saying he hopes the technology absorbs routine, unglamorous work so doctors can spend more time with patients.

Why it matters

If the findings survive clinical validation, diagnostic medicine would become one of the first domains where AI demonstrably outperforms trained professionals rather than merely assisting them — a meaningful shift given how persistent misdiagnosis remains. The architectural lesson may travel even further than the medicine: Microsoft's edge came from orchestrating several agents rather than waiting for a single smarter model, an approach the company argues retains its value as base models improve. For now, though, the evidence is retrospective. The distance between an 83 percent score on archived case reports and a tool that hospitals, regulators, and doctors trust at a bedside is exactly the leap Microsoft admits it has not yet taken.