Research
Microsoft's AI gets more right than general doctors on specialist cases
A Microsoft study in Science claims its system diagnoses difficult cases far better than doctors. The catch: it was a simulation, and it didn't compare the system with specialists, who are the ones who handle those cases.

The essentials
- Claimed by one partyAccording to Microsoft's study in Science, its MAI-DxO system got 83% of complex cases right, compared with 36% for general doctors (TechCrunch).
- Claimed by one partyThe comparison left out specialists and medical teams, who usually handle cases like these, according to TechCrunch (TechCrunch).
- Claimed by one partyAccording to TechCrunch, much of the improvement comes from the underlying AI model. The best results came with OpenAI's GPT-5 and GPT-5.5 (TechCrunch).
- Claimed by one partyMicrosoft will test the system with real patients who give their consent at Beth Israel Deaconess Medical Center in Boston (TechCrunch).
Why it matters
Microsoft joins Google and OpenAI in saying its AI matches or beats doctors (TechCrunch). But the test is a simulation designed by the company itself, against doctors who don't usually see cases like these. The data suggest that what improves most is the base model, not Microsoft's system. The real test will be with patients in Boston.
The details
According to TechCrunch, MAI-DxO coordinates several AI agents (programs that act on their own) as if they were a panel of doctors. They question the patient, order tests and debate the diagnosis (TechCrunch). Microsoft gathered the cases into its own test and, according to The Verge, talks of up to four times more accuracy than doctors (The Verge). Mustafa Suleyman, its head of AI, sees it as a step toward "medical superintelligence" (TechCrunch).
According to TechCrunch, Microsoft's system improved GPT-5 only slightly on complex cases and GPT-5.5 quite a bit more, taking it from 72% to almost 87% correct. On emergency cases it made almost no difference with either model (TechCrunch). It helps most with less capable models, although the gap between one model and another matters more than that boost (TechCrunch).
What we don't know
- How the system would do against specialists or medical teams, who weren't part of the comparison.
- When the study with real patients in Boston starts and how long it will last.
- Whether anyone outside Microsoft has repeated the test.
Sources
- Microsoft says its AI tool is better than doctors at diagnosing conditions
- Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’
- Can AI diagnose like a team of doctors? New research puts it to the test
Was this story useful? Yes Not really