Microsoft AI Diagnoses Complex Cases 83% of the Time

Microsoft’s new diagnostic AI system correctly identified complex medical conditions in a simulation 83% of the time, more than double the accuracy of generalist physicians. The study, published in Science on Thursday, marks a significant milestone for Microsoft AI as it pursues what CEO Mustafa Suleyman describes as 'medical superintelligence.' The system, known as MAI-DxO, functions by coordinating multiple AI agents that simulate a panel of doctors. These agents question patients, order tests, and debate diagnoses to reach a conclusion.

The benchmark test used complex cases from the New England Journal of Medicine. While the AI achieved an 83% accuracy rate, primary care doctors and general internists working individually with reference materials were correct only 36% of the time. It is important to note that these specific cases are typically handled by specialists or teams in real-world settings, who were not part of this direct comparison. Microsoft joins Google and OpenAI in asserting that its AI can match or exceed human performance, though the data suggests much of the improvement stems from underlying model power rather than just the orchestration layer.

Performance varied depending on the underlying model used. The best results were achieved with OpenAI’s GPT-5 and GPT-5.5. For GPT-5, the DxO system reduced the estimated cost per case by approximately $3,000 while improving accuracy by about four points. However, on GPT-5.5, a model released in April 2026, the system boosted accuracy from 72% to nearly 87% at a slightly lower cost. In contrast, emergency department cases showed little improvement in either accuracy or cost whether the AI used the orchestration system or not. Weaker models like Gemini 2.5 Flash saw a substantial jump from 46.4% to nearly 69% with the help of DxO, though the gap between weaker and stronger models remained larger than the boost provided by the system.

Reactions to the findings emphasize the potential to address the growing mismatch between medical demand and supply. Dominic King, vice president of Health at Microsoft AI, stated there is 'something very substantial and exciting here.' He noted that the technology is ready for real-world evaluations, which will begin with a new partnership with Beth Israel Deaconess Medical Center in Boston. This study will compare DxO’s recommendations against human doctors in both emergency and urgent care settings, aiming to provide compelling evidence for broader deployment.

Suleyman, who previously co-founded DeepMind, believes that within the next 24 to 36 months, healthcare models will see an order of magnitude improvement in capability. He is particularly excited about a potential breakthrough in 2027 for 'coding style' medical records, which would allow clinicians to stream highly accurate predictions about patient outcomes. Microsoft is also working with the Mayo Clinic to train a foundation model for healthcare, with a release planned for next year. The company views this as a differentiator against other frontier labs, aiming to become the leader in state-of-the-art clinical decision support products.

The publication of this study coincides with a manifesto defining medical superintelligence as the capacity to surpass the strongest contemporary human-led health systems. Microsoft hopes to move beyond simple benchmarking tasks like answering exam questions into deeper integration with clinicians and healthcare systems. As the company already sells dictation and radiology tools, DxO represents an expansion of its existing efforts. Suleyman remains confident that this focus will allow Microsoft to accelerate the introduction of this technology to the world, citing the rapid progress seen in other AI fields as a sign of what is to come.