Research regarding the diagnostic tool was published in the scientific journal Science on 22 October. In complex medical cases, the AI achieved approximately 83% accuracy, while doctors using reference materials reached only 36% accuracy.
Physicians working without reference materials managed 22% accuracy in those same complex scenarios. The MAI-DxO was roughly 3.7 times more accurate than doctors without references and 2.3 times more accurate than those using reference materials.
Emergency department cases showed an 80% accuracy rate for the AI, compared to 47% for human physicians. The system utilizes OpenAI’s GPT-5 model to process medical data.
Microsoft developed a testing framework called Sequential Diagnosis Benchmark (SDBench) to evaluate the technology. Testing involved 304 complex medical cases from The New England Journal of Medicine and 325 emergency department cases from Beth Israel Deaconess Medical Center in Boston.
The company aims to develop "medical superintelligence" but has not announced a commercial availability date for the Diagnostic Orchestrator.
