AI News · Hot news ·
Microsoft says its MAI-DxO system solved 83% of hard cases versus 36% for doctors
Microsoft's MAI-DxO diagnostic system, described in a study published in Science on October 22, was correct on 83% of complex New England Journal of Medicine cases in a simulation. Primary care doctors and general internists working alone were correct 36% of the time. Microsoft AI CEO Mustafa Suleyman called the result a step toward "medical superintelligence."
Key points
- MAI-DxO was correct on 83% of complex simulated cases versus 36% for generalist doctors.
- With GPT-5, the system cut cost per case by about $3,000 and added about four points.
- A real world evaluation with Beth Israel Deaconess Medical Center is planned.
What happened: Microsoft's MAI-DxO system was described in a study published Thursday in Science, TechCrunch reported. The system coordinates several AI agents that act like a panel of doctors, questioning a patient, ordering tests and debating a diagnosis. In a simulation using complex New England Journal of Medicine cases, it was correct 83% of the time, compared with 36% for primary care doctors and general internists working alone with reference materials. Specialists and physician teams were not part of the comparison.
The details: Microsoft tested MAI-DxO with models from OpenAI, Google, Anthropic, Meta and DeepSeek, with the best results from GPT-5 and GPT-5.5. GPT-5.5 rose from 72% to nearly 87% on complex cases with the system, at slightly lower cost. GPT-5 alone scored 78.9% at about $8,400 per case; with MAI-DxO, cost fell by about $3,000 and accuracy rose about four points. On emergency department cases, the system made little difference. Microsoft researchers said gains shrink as models approach the theoretical maximum.
What to watch: Microsoft will evaluate the system with Beth Israel Deaconess Medical Center in Boston, comparing its recommendations with those of human doctors for consenting patients. It is also working with Mayo Clinic on a healthcare foundation model planned for next year. Suleyman said he expects healthcare models to get "an order of magnitude better in capability" within 24 to 36 months.
Our take
The study suggests orchestrating several model agents can raise accuracy and cut cost per task, a pattern relevant to teams building agent workflows well beyond healthcare.