Microsoft AI Beats Doctors on Hard Cases in New Study

For patients facing difficult medical puzzles, a new AI system from Microsoft shows promise. In a simulation published in Science, the technology correctly diagnosed complex cases more than twice as often as generalist physicians. This result marks a significant shift in how the tech giant views its role in healthcare, moving beyond simple tools to more complex decision support.

The system, known as MAI-DxO, works like a panel of doctors. It uses several AI agents to question patients, order tests, and debate diagnoses. In tests using complex cases from the New England Journal of Medicine, the system was correct 83% of the time. Generalist doctors and primary care physicians, working alone with reference materials, were correct only 36% of the time. It is important to note that these difficult cases are usually handled by specialists or teams, who were not part of this specific comparison.

Microsoft AI CEO Mustafa Suleyman called the result a step toward medical superintelligence. He told TechCrunch that the company is working with the Mayo Clinic to train a foundation model for healthcare. This new model is planned for release next year. Suleyman believes this infrastructure will allow for real breakthroughs in clinical decision support.

The study also looked at costs. Using OpenAI’s GPT-5 model, the system cost about $8,400 per complex case and scored 78.9%. The MAI-DxO system reduced that cost by about $3,000 but only raised accuracy by four points. With the newer GPT-5.5 model, results varied. On complex cases, accuracy jumped from 72% to nearly 87% with the system's help. However, on emergency department cases, there was little difference in accuracy or cost between using the model alone and using it with MAI-DxO.

The technology helps most with weaker AI models. For example, a model called Gemini 2.5 Flash scored 46.4% on its own but nearly 69% with the system. Despite this boost, the gap between weaker and stronger models remained larger than the improvement provided by the orchestration system. Researchers noted that gains naturally shrink as models approach their theoretical maximum.

Microsoft is moving this technology into real hospitals. Through a partnership with Beth Israel Deaconess Medical Center in Boston, the company will evaluate the system across different settings. Patients will receive standard care. Those who consent will also engage with the AI. Doctors and researchers will then compare the AI's recommendations against human decisions. This study aims to provide compelling evidence for scaling these systems in the future.

This launch comes as Microsoft releases a manifesto on medical superintelligence. The company defines this as the ability to surpass human-led health systems in reasoning and coordination. Suleyman said he expects healthcare models to become an order of magnitude better within the next 24 to 36 months. He pointed to coding and chatbots as examples of how progress accelerates once a model is in production and being iterated on with real-world usage.

Suleyman has long focused on using AI for medical issues. He was a co-founder of DeepMind before joining Microsoft in 2024. He sees this new focus as a way to differentiate Microsoft from other AI labs. The company already sells dictation and radiology tools to clinicians. With this new system, it aims to become the top provider of medical clinical decision support products in the coming years.