Microsoft claims AI diagnostic system outperforms generalist physicians in complex cases

83 percent. That is the accuracy rate Microsoft’s new AI diagnostic system, MAI-DxO, achieved in a simulation of complex medical cases, a figure more than double the 36 percent success rate recorded by generalist physicians working individually. The study, published in Science on Thursday, positions the technology as a significant step toward what Microsoft AI CEO Mustafa Suleyman calls “medical superintelligence,” a term he uses to describe a capacity that surpasses current human-led health systems in reasoning and coordination.

The system operates by coordinating multiple AI agents that function similarly to a panel of specialists. These agents interact with the patient, order necessary tests, and debate potential diagnoses before reaching a conclusion. The test data was drawn from complex cases originally published in the New England Journal of Medicine. However, the comparison group consisted of primary care doctors and general internists who used reference materials but worked alone. The study notes that such complex cases are typically handled by specialists or teams in real-world settings, a nuance that complicates the direct comparison to solo generalists.

Dominic King, Vice President of Health at Microsoft AI, characterized the results as substantial, citing a growing mismatch between the demand for and supply of medical expertise. By releasing these findings, Microsoft joins Google and OpenAI in asserting that its artificial intelligence can match or exceed human diagnostic capabilities. The data suggests that much of the performance gain stems from the power of the underlying large language models, with the orchestration system adding incremental improvements rather than creating them from scratch.

Microsoft tested the system using models from several competitors, including Anthropic, Meta, DeepSeek, and Google, but found the highest performance with OpenAI’s GPT-5 and GPT-5.5. Cost analysis revealed that while the orchestration system reduced the estimated cost per case by approximately $3,000 compared to using GPT-5 alone, the accuracy gain was modest, increasing by only about four points to 78.9 percent. In emergency department scenarios, the system offered little improvement in either cost or accuracy over the standalone model.

With the newer GPT-5.5 model, released in April 2026, the results were more varied. For complex cases, the system raised accuracy from 72 percent to nearly 87 percent at a slightly lower cost. Yet, in emergency settings, both the assisted and unassisted models scored around 85 percent with similar costs. The orchestration system proved most beneficial for less capable models; for instance, Gemini 2.5 Flash saw its accuracy jump from 46.4 percent to nearly 69 percent when paired with the system, though the gap between strong and weak models remained wider than the boost provided.

Suleyman, who joined Microsoft in 2024 to lead its AI efforts after co-founding DeepMind, stated the company is collaborating with the Mayo Clinic to train a healthcare foundation model, planned for release next year. He predicts that by 2027, the technology will enable “streaming” predictions of how medical records might unfold, offering clinicians foresight he describes as “gold dust.” He expects healthcare models to improve by an order of magnitude in capability within 24 to 36 months, driven by a “flywheel” of real-world usage and iteration.

To validate these capabilities, Microsoft is partnering with Beth Israel Deaconess Medical Center in Boston to evaluate the system in real-world hospital settings, including emergency and urgent care environments. Patients will receive standard treatment while consenting to parallel evaluation by the AI. King described the initiative as potentially the largest study of its kind, aiming to provide compelling evidence for broader deployment. The study coincides with a manifesto outlining the path to medical superintelligence, shifting focus from isolated tasks to integrated care.

Microsoft currently sells dictation and radiology tools to clinicians and views this diagnostic system as an expansion of that portfolio. Suleyman emphasized that the company aims to have the number one state-of-the-art medical clinical decision support product in the coming years, viewing this focus as a key differentiator from other frontier labs. He expressed confidence that the technology is ready for real-world scrutiny, noting the complexity of issues that arise in settings like a Boston emergency department on a Friday evening.