Microsoft MAI-DxO published in Science with 83% accuracy on complex cases versus 36% for physicians
Microsoft published MAI-DxO in Science, reaching 83% accuracy on NEJM complex cases versus 36% for physicians. The system uses GPT-5.5 as its reasoning engine at a reported inference cost of about $8,400 per case.
Microsoft chief research lead Mustafa Suleyman said on October 22 that his team published a paper in Science introducing a diagnostic system called MAI-DxO. On difficult cases from the New England Journal of Medicine the system reached 83% accuracy against 36% for the physician comparison group. On emergency department cases it scored around 80% versus 47% for physicians.
The system uses GPT-5.5 as its reasoning engine. GPT-5.5 answering alone scored 72%, rising to 87% when connected to the DxO diagnostic orchestration framework. With Gemini 2.5 Flash the figure moved from 46.4% to 69%. Microsoft also disclosed that GPT-5 alone costs roughly $8,400 per case to reason through.
Suleyman framed the goal as medical superintelligence, with a path involving a dedicated medical model delivered with Mayo Clinic in 2027. Microsoft also announced a partnership with Beth Israel Deaconess Medical Center. The paper notes the system still struggles clearly on rare diseases and multi-morbidity cases, and that effects on uneven medical resource distribution need further evaluation.