Microsoft’s AI diagnostic system, known as MAI-DxO, has demonstrated a remarkable ability to outperform generalist doctors on complex medical cases in simulations. According to a study published in Science, MAI-DxO accurately diagnosed difficult cases 83% of the time, more than twice the success rate of individual primary care physicians, who scored 36%. This system operates by coordinating multiple AI agents that mimic a panel of doctors through patient questioning, test ordering, and collaborative diagnosis. While these complex cases typically require specialist teams, the study highlights a strong potential for AI to augment medical decision-making.

The improved performance of MAI-DxO largely stems from the underlying AI models it integrates rather than the orchestration system itself. Microsoft tested the system using various proprietary models, including OpenAI’s GPT-5 and GPT-5.5, Google’s, Anthropic’s, and others, finding its best results with OpenAI’s latest iterations. For example, the DxO system raised accuracy on the GPT-5.5 model from 72% to nearly 87% for complex cases while also reducing costs slightly. Despite these gains, its advantage diminishes with emergency department cases, where performance remains similar with or without DxO.

Mustafa Suleyman, CEO of Microsoft AI and former DeepMind co-founder, aims to push towards what he terms “medical superintelligence,” where AI surpasses the best human-led health systems in diagnostic reasoning and care coordination. Microsoft is collaborating with the Mayo Clinic to develop a foundation medical AI model expected to launch next year, which Suleyman believes will enable breakthroughs in clinical decision support and predictive insights from medical records. The company is also partnering with Beth Israel Deaconess Medical Center to test MAI-DxO in real hospital settings, comparing AI-driven recommendations with those from human physicians in urgent care environments.

Microsoft views AI-driven medical tools as a major frontier for its technology, complementing existing products like dictation and radiology assistance. Suleyman emphasizes that while the integration of AI into healthcare is still early, rapid progress is anticipated in the coming years due to iterative improvements driven by real-world use data. The research and upcoming real-world evaluations signify Microsoft’s commitment to accelerating the deployment of advanced clinical decision support systems, differentiating itself in the competitive AI healthcare landscape dominated by other major players like Google and OpenAI.