Key Takeaways
- MAI-DxO accuracy on NEJM cases: 83%, against 36% for generalist doctors using reference materials.
- Simulated cases: the figures describe evaluated tasks, not outcomes from patients receiving AI-directed care.
- Gap over physicians: 47 percentage points with reference materials and 61 without them.
- Emergency-department cases: MAI-DxO scored 80% against 47% for physicians.
Microsoft AI reported on 22 October 2026 that its MAI Diagnostic Orchestrator, or MAI-DxO, correctly diagnosed 83% of complex simulated cases drawn from the New England Journal of Medicine (NEJM) in a study published that day in Science. Generalist doctors who had reference materials scored 36% on the same cases. Technology.org reports that the work first appeared in mid-2025, when Microsoft described an earlier version of MAI-DxO.
Microsoft's post gives 22% for physicians without external reference resources and 36% with them, a 14-point gain from the materials. MAI-DxO's 83% sits 47 points above the 36% and 61 points above the 22%. By WYDE's arithmetic from Microsoft's figures, the 47-point margin is more than three times that 14-point gain.
Simulated Cases and Generalist Doctors: What the Results Cover
The results come from simulations, not from clinical care. Technology.org reports that MAI-DxO solved 83% of complex simulated cases, while generalist physicians working alone with reference materials solved 36%. The physicians in that comparison were generalists, meaning primary care doctors and general internists. The outlet describes the paper as Microsoft's most ambitious medical AI claim, now in a peer-reviewed journal.
Microsoft's own post says benchmarks matter, but the true test is performance in real clinical environments. A write-up on remio.ai says the figures describe evaluated tasks rather than outcomes from patients receiving AI-directed care. It adds that the study does not, by itself, establish that the system is available for routine clinical deployment.
No source reports that MAI-DxO is approved or cleared for clinical use.
MAI-DxO Accuracy on NEJM and Emergency-Department Cases
On complex NEJM cases, MAI-DxO achieved 83% diagnostic accuracy. Physicians achieved 22% without external reference resources and 36% with them. Microsoft says that makes MAI-DxO about 3.7 and 2.3 times more accurate, respectively.
On more typical emergency-department cases, MAI-DxO achieved 80% accuracy against 47% for physicians. Microsoft describes that as more than 1.7 times higher. The gap on those cases is 33 percentage points. Technology.org says the Science paper broadens the evidence with emergency-department cases.
Microsoft also reports a cost-focused tuning. When MAI-DxO was optimized more heavily for cost, its performance remained strong while it chose more cost-effective workups. One configuration reached 81% accuracy on NEJM cases at less than half the cost of comparative systems. The sources give no dollar figure for that comparison.
What AI Diagnosis Could Mean for Patients and Doctors
For patients, the point is access, as King frames it. Dominic King, vice president of Health at Microsoft AI, said: "The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise." The study does not test whether a system like this would narrow that gap in care, because it scores accuracy on simulated cases.
For doctors, the study measures generalists working alone, with or without reference materials, on simulated cases, not a workflow in a clinic. King also said: "There is something very substantial and exciting here," and Technology.org reports that outside physicians keep pressing for trials with real patients. It also notes that Microsoft now joins Google and OpenAI in claiming AI that can match or beat doctors on diagnosis.
Related WYDE coverage: Nature Medicine Stress Test Finds Frontier AI Models Not Clinically Ready, and The FDA Opens a Public Docket on Generative AI Medical Devices and Floats Testing Them the Way Doctors Are Tested.
People Also Ask
What did Microsoft's MAI-DxO score in the Science study?
MAI-DxO diagnosed 83% of complex simulated NEJM cases in the Science study, Microsoft reported. Physicians scored 36% with reference materials and 22% without them, according to the same post.
Was MAI-DxO tested on real patients?
The reported results come from simulated cases, not from patients receiving AI-directed care. Microsoft says benchmarks matter, but the true test is performance in real clinical environments, and none of the sources reports patient outcomes for MAI-DxO.
How much cheaper was the cost-focused version of MAI-DxO?
Microsoft said one configuration reached 81% accuracy on NEJM cases at less than half the cost of comparative systems. The sources give no dollar figure, so the exact size of that saving is not known from them.
Which doctors were compared with MAI-DxO?
The comparison group was generalist doctors, meaning primary care doctors and general internists, according to Technology.org. Those comparisons used simulated cases, so they do not show how the same doctors perform in clinics.
Sources: Microsoft AI, Technology.org, Remio.