Microsoft system diagnosed 83% of complex cases in simulation — TechCrunch
Microsoft’s MAI-DxO artificial intelligence diagnostic system correctly identified 83% of complex medical cases in a simulation, compared with 36% for general practitioners and internists working individually with reference materials. TechCrunch reports on the results of the study published in the journal Science.
MAI-DxO coordinates several AI agents that simulate the work of a medical consultation: they ask the patient questions, order tests and discuss a possible diagnosis. Complex clinical cases from the New England Journal of Medicine were used for testing.
Limitations of the comparison
Such complex cases are usually reviewed by specialized experts or medical teams, who were not included in the comparison with the system. Microsoft AI CEO Mustafa Suleyman called the result a step toward “medical superintelligence.”
More current news is available on the UA.News Telegram channel Telegram.
At the same time, the study’s own data indicate that basic language models account for a significant portion of the accuracy gain, while the DxO coordination system adds a separate effect. Microsoft tested MAI-DxO on models from OpenAI, Google, Anthropic, Meta and DeepSeek; the best results were achieved with OpenAI’s GPT-5 and GPT-5.5.
Accuracy and cost
In tests of complex cases, GPT-5 with a basic short prompt demonstrated 78.9% accuracy at an estimated cost of $8,400 per case. Using DxO reduced the estimated cost by about $3,000, but increased accuracy by only about four percentage points. In scenarios for emergency departments, no significant improvement in accuracy or cost was recorded.
DxO most noticeably helped less powerful models: Gemini 2.5 Flash achieved 46.4% accuracy on its own and nearly 69% with the coordination system. Microsoft also plans to assess DxO’s clinical reasoning in hospital settings together with Beth Israel Deaconess Medical Center in Boston. Patients who provide consent will interact with the system alongside standard treatment, and its recommendations will be compared with doctors’ decisions.
