🏥Microsoft MAI-DxO Hits 83% on Hard Diagnosis Cases
TL;DR
Microsoft AI's multi-agent MAI-DxO scored 83% on complex NEJM cases in simulation, versus 36% for primary care doctors working alone. The Science-published study excludes specialists and teams, the people who usually handle such cases.
Microsoft AI's multi-agent MAI-DxO scored 83% on complex NEJM cases in simulation, versus 36% for primary care doctors working alone. The Science-published study excludes specialists and teams, the people who usually handle such cases.
Key Points
DxO orchestrates several agents that question patients, order tests and debate diagnoses
GPT-5.5 with DxO rose from 72% to nearly 87% on complex cases at slightly lower cost
Gemini 2.5 Flash jumped from 46.4% to nearly 69% with DxO
Little lift in emergency cases: both setups scored about 85%
Next steps: a Mayo Clinic foundation model next year and a real-world comparison with Beth Israel Deaconess
Why It Matters
Most of the gain still comes from the base model, so the orchestration layer helps weaker models most. Claims of 'medical superintelligence' need specialist baselines first.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.