Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’

Your medical diagnostic support tool is more accurate than a human general practitioner on complex cases.

Microsoft's AI diagnostic system, MAI-DxO, demonstrated a significant advancement in medical accuracy during a simulation, outperforming generalist physicians. In a study published in Science, MAI-DxO correctly diagnosed complex medical cases 83% of the time, compared to 36% for human generalists using reference materials. The AI system, which coordinates multiple AI agents to simulate a panel of doctors, was tested on complex cases from the New England Journal of Medicine. While human specialists or teams typically handle such complex cases, this study focused on individual generalists. Microsoft AI CEO Mustafa Suleyman highlighted this as a step towards 'medical superintelligence,' aiming to address the mismatch between demand and supply of medical expertise. The company is collaborating with the Mayo Clinic to train a healthcare-specific foundation model, with a planned release next year, and is initiating real-world evaluations with Beth Israel Deaconess Medical Center.

Microsoft's MAI-DxO system, which leverages an orchestration of multiple AI agents, showed varying degrees of improvement depending on the underlying AI model. It significantly boosted accuracy for weaker models like Gemini 2.5 Flash, raising its performance from 46.4% to nearly 69%. However, for advanced models like GPT-5.5, the gains were less pronounced on emergency department cases, suggesting that performance plateaus with highly capable models. The system's primary benefit appears to be on complex, rather than routine, medical scenarios, and it also impacts cost efficiency, though the exact savings vary.

This development signals a growing trend of AI systems not just mimicking, but actively seeking to surpass human capabilities in specialized fields, moving towards 'medical superintelligence.' Microsoft's strategy of integrating AI agents and developing specialized foundation models for healthcare, in partnership with institutions like the Mayo Clinic and Beth Israel Deaconess, indicates a clear path toward clinical deployment. The company's ambition is to move beyond task-specific AI to systems that can reason and coordinate across the full spectrum of patient care, potentially reshaping clinical decision-making and healthcare delivery.