Microsoft’s Sequential Diagnosis with Language Models Outperform Physicians
In brief
Microsoft's MAI-DxO achieved higher accuracy than generalist physicians on an interactive diagnostic benchmark, reaching 83.2% and 80.3% versus 36.1% and 46.8% for the physicians. The system starts from the chief complaint, asks follow-up questions, selects tests, and evaluates diagnostic cost. The paper was published in Science on 2026-10-22, DOI 10.1126/science.aeh7930.
Why it matters
Comparison data lists model and physician accuracy on two case types, with discussion of benchmark realism and low human baselines.
