83%: The AI diagnostic score that outpaced human doctors

For patients facing difficult medical diagnoses, the gap between human and artificial intelligence assessment just widened. A new study published Thursday in Science reveals that Microsoft’s AI diagnostic system, MAI-DxO, correctly identified complex medical cases more than twice as often as generalist physicians in a controlled simulation. While the technology is not yet in standard clinical use, these results mark a significant milestone in the race to integrate advanced AI into everyday healthcare delivery.

The system operates by coordinating multiple AI agents that function as a collaborative medical panel. These agents simulate a consultative process by questioning virtual patients, ordering tests, and debating potential diagnoses. In simulations using complex cases from the New England Journal of Medicine, the AI achieved an 83% accuracy rate. By comparison, primary care doctors and general internists, working individually with reference materials, were correct only 36% of the time. It is important to note that these specific cases are typically handled by specialists or teams in real-world settings, who were not included in this direct comparison.

Microsoft attributes much of the performance boost to the underlying power of the AI models rather than just the orchestration system. The company tested the framework with models from OpenAI, Google, Anthropic, Meta, and DeepSeek, finding the best results with OpenAI’s GPT-5 and GPT-5.5. When using GPT-5.5, the AI system increased accuracy from 72% to nearly 87% on complex cases while slightly reducing costs. However, the performance gains were less pronounced for emergency department cases, where the AI and the model alone performed similarly.

The study also highlights the economic efficiency of the technology. While running GPT-5 alone on complex cases cost approximately $8,400 per case with a 78.9% accuracy rate, the MAI-DxO system reduced that cost by about $3,000 while raising accuracy by four points. This cost-accuracy optimization is a key focus for Microsoft, which aims to make advanced diagnostic support more accessible without significantly increasing healthcare expenditures for providers.

Microsoft is moving quickly to validate these lab results in real-world environments. The company is partnering with the Mayo Clinic to train a specialized healthcare foundation model, with a release planned for next year. Additionally, a new collaboration with Beth Israel Deaconess Medical Center in Boston will evaluate the system’s clinical reasoning across various hospital settings. Patients receiving standard care will be offered the parallel use of the AI, allowing researchers to compare its recommendations against those of human doctors in high-pressure environments like emergency rooms.

Mustafa Suleyman, CEO of Microsoft AI, described the current work as a step toward "medical superintelligence," defined as the capacity to surpass contemporary human-led health systems in reasoning and coordination. He predicts that within 24 to 36 months, healthcare models will improve by an order of magnitude, driven by a "flywheel" of real-world usage and iterative improvements. This approach mirrors the rapid advancements seen in AI coding and chatbots, suggesting that the next major breakthrough in medical records and clinical decision support could arrive as early as 2027.