83 percent on complex cases, four times is the lede

This Verge page is stamped Oct 22, 2026, 6:00 PM UTC. Microsoft says the Diagnostic Orchestrator results were published Thursday in Science. The lede says the company claims the tool can diagnose conditions up to four times more accurately than human doctors. Four times is that lede. The same page later prints a different set of percentages. This story does not rewrite four times as 83 percent versus 36 percent.

On case reports from The New England Journal of Medicine, the system reached about 83 percent, compared with about 36 percent for generalist physicians using references and 22 percent for those working without them. On emergency admissions from Beth Israel Deaconess Medical Center in Boston, the system scored 80 percent against an average of 47 percent for human physicians. Microsoft compiled the cases into its own benchmark, SDBench, which it plans to open-source. The plan is not a release. The researchers say the system is powered by OpenAI's GPT-5 and also beat frontier models running alone. The peer-reviewed paper does not include newer models. The page names GPT-6 Astra, Meta's Muse Wonder Full, Google's Gemini 4 Argon, and Anthropic's Opus 5.5 as not included. The system tended to reach a diagnosis at a lower estimated cost. This page does not say half or one third.

The tests used only past cases. Suleyman says it needs clinical validation. King tells The Verge that the company plans to begin hospital testing with that medical center. Neither would commit to a firm date. King says it will likely get underway in the next couple of months. Microsoft has yet to say when or how the technology might become available.

A second page, from IT Home, is stamped 2026/10/23 7:23:36. It describes the same system, named MAI-DxO, short for DxO, and gives the full name Microsoft AI Diagnostic Orchestrator. On common emergency visits in the MIMIC-IV-ED set, it says diagnostic accuracy reached 1.7 times that of human doctors. The 1.7 times is this page's figure. It does not print 80 percent or 47 percent, so those Verge figures stay a separate score. On complex cases, this page prints 83 percent, against 36 percent for general practitioners and general internists who could consult references on their own. It does not print 22 percent. It also says these complex cases are usually handled by specialists or a multidisciplinary team, and that those specialists were not in the comparison.

Cost is not the same sentence on the two pages. IT Home says the system costs half or even a third of a doctor. The Verge only says the estimated cost tended to be lower. IT Home says the brightest models among those tested were GPT-5 and GPT-5.5. The Verge says the system is powered by GPT-5 and names four newer models the paper does not include. Both pages point to Science. IT Home prints the paper link. The paper itself was not read beyond these two pages. This story keeps the published accuracy, who was in the comparison, and each page's cost wording. It does not describe how a diagnosis or a treatment is carried out.