Healthcare AI: Microsoft claims its diagnostic system beats human doctors on complex cases

For patients facing difficult medical puzzles, the gap between available expertise and demand is widening. Microsoft argues its new AI diagnostic system, MAI-DxO, can help close that divide. In a simulation published this week in the journal Science, the system correctly diagnosed complex cases more than twice as often as generalist physicians, a result the company’s CEO, Mustafa Suleyman, describes as a step toward “medical superintelligence.”

The system operates by coordinating multiple AI agents that function like a medical panel. These agents question the patient, order tests, and debate potential diagnoses before reaching a conclusion. In the study, which used complex cases from the New England Journal of Medicine, the AI was correct 83% of the time. By comparison, primary care doctors and general internists working individually with reference materials were correct only 36% of the time. However, the study notes that such complex cases are typically handled by specialists or teams, who were not included in this specific comparison.

Dominic King, vice president of Health at Microsoft AI, emphasized the urgency of the problem. “The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise,” King told TechCrunch. By releasing these results, Microsoft joins Google and OpenAI in asserting that their AI systems can match or exceed human performance in medical diagnostics.

The data suggests that much of the improvement stems from the power of the underlying large language models, with Microsoft’s orchestration layer adding incremental gains. The best performance was achieved using OpenAI’s GPT-5 and GPT-5.5 models. When using GPT-5.5, the orchestration system raised accuracy from 72% to nearly 87% on complex cases at a slightly lower cost. However, the benefits were less pronounced for emergency department cases, where the AI performed similarly to the base model. Additionally, the system provided a significant boost to weaker models, such as Gemini 2.5 Flash, which improved from 46.4% to nearly 69% with the assistance of the diagnostic framework.

Microsoft is now moving from simulation to real-world testing. The company has partnered with the Mayo Clinic to train a foundation model for healthcare, with a release planned for next year. Suleyman believes this will allow for “really enormous leaps” in the field. He also noted that Microsoft is working with Beth Israel Deaconess Medical Center in Boston to evaluate the system’s clinical reasoning across various hospital settings, including urgent care and emergency rooms.

In this upcoming study, patients will receive standard treatment while consenting to parallel evaluation by the AI. The goal is to compare the AI’s recommendations against human decisions to gather compelling evidence for wider deployment. Suleyman expects healthcare models to improve by an order of magnitude within the next 24 to 36 months, driven by a “flywheel” of real-world usage and iteration. He views this focus on medical records and clinical decision support as a key differentiator for Microsoft in the competitive landscape of frontier AI labs.