Microsoft claims its AI diagnostic panel outperforms general physicians

The race for artificial intelligence to handle complex healthcare tasks has intensified as major tech firms compete to prove their models can match human expertise. On Thursday, Microsoft released findings showing that its new diagnostic system, MAI-DxO, successfully identified correct diagnoses in difficult cases more than twice as often as generalist doctors.

This development follows a rapid timeline of AI integration in medicine. Mustafa Suleyman, the CEO of Microsoft AI, has led efforts to apply frontier AI to clinical problems since joining the company in 2024. Previously, he co-founded DeepMind and launched health initiatives with Google. The current push builds on these foundations, aiming to move beyond simple chatbots to systems that can coordinate complex medical reasoning.

The system works by coordinating multiple AI agents that function like a panel of specialists. These agents simulate a medical team by questioning a virtual patient, ordering tests, and debating potential diagnoses. In simulations using complex cases from the New England Journal of Medicine, the system achieved an accuracy rate of 83%. By comparison, primary care doctors and general internists working individually with reference materials were correct only 36% of the time. It is worth noting that such complex cases are typically managed by specialists or teams, who were not included in this specific comparison.

The results suggest that much of the performance gain comes from using powerful underlying models, with Microsoft’s orchestration adding incremental benefits. The system performed best when paired with OpenAI’s GPT-5 and GPT-5.5 models. For instance, using GPT-5.5, the system raised accuracy on complex cases from 72% to nearly 87% while slightly reducing costs. However, the advantage was less clear for emergency department cases, where both the AI and human doctors showed similar performance levels.

Looking ahead, Microsoft plans to test the system in real-world settings. Through a partnership with Beth Israel Deaconess Medical Center in Boston, the company will evaluate the AI’s clinical reasoning across various hospital environments. Patients will continue to receive standard care, while those who consent will also interact with the AI system. Researchers will then compare the AI’s recommendations against the decisions made by human physicians. Suleyman believes this approach will lead to significant breakthroughs in medical record prediction by 2027.

Microsoft also published a manifesto defining its goal of achieving medical superintelligence, which it describes as surpassing human-led systems in reasoning and coordination. The company aims to integrate these tools into broader healthcare workflows rather than just benchmarking isolated tasks. Suleyman stated that the company expects healthcare models to become significantly more capable within the next two to three years as they are refined through real-world usage.