Microsoft's AI Panel Outpaces Doctors in Complex Medical Diagnoses

For patients facing difficult medical symptoms, the landscape of diagnostic support is shifting. A new system developed by Microsoft has shown the ability to correctly identify complex medical conditions far more often than experienced generalist physicians. This development suggests that artificial intelligence is moving from a supportive tool to a primary diagnostic partner in scenarios where human judgment often struggles.

The system, known as MAI-DxO or DxO, operates like a virtual panel of specialists. Instead of relying on a single algorithm, it coordinates multiple AI agents to question the patient, order necessary tests, and debate potential diagnoses. In a simulation using difficult cases from the New England Journal of Medicine, this collaborative approach proved highly effective.

The accuracy gap between the AI and human doctors was substantial. DxO arrived at the correct diagnosis in 83% of the complex cases. In contrast, primary care doctors and general internists, who worked individually with reference materials, were correct in only 36% of instances. It is important to note that these specific cases are typically managed by specialist teams in real-world settings, so the comparison focuses on generalist capabilities.

The performance of the system depends heavily on the underlying AI model it uses. Microsoft tested DxO with models from various providers, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The highest results were achieved using OpenAI’s GPT-5 and GPT-5.5. The study indicates that much of the improvement comes from the power of these base models, with the DxO orchestration adding additional gains on top.

Cost and efficiency were also part of the evaluation. When using GPT-5 alone, the system cost about $8,400 per case and achieved 78.9% accuracy. Adding the DxO orchestration reduced the cost by roughly $3,000 while increasing accuracy by four points. However, for emergency department cases, the added complexity of DxO provided little benefit in either cost or accuracy compared to using GPT-5 on its own.

The technology shows the most promise when paired with less powerful models or when dealing with intricate diagnoses. For example, Gemini 2.5 Flash scored 46.4% independently but rose to nearly 69% with the assistance of DxO. Despite this boost, the gap between stronger and weaker base models remained larger than the specific improvement provided by the orchestration system.

Microsoft is now moving toward real-world validation through a partnership with Beth Israel Deaconess Medical Center in Boston. This evaluation will test DxO across various hospital settings, including urgent care and emergency rooms. Patients will receive standard treatment while consenting to engage with the AI in parallel. The goal is to compare the AI’s recommendations directly against those of human clinicians.

Mustafa Suleyman, CEO of Microsoft AI, views this as a step toward what he calls medical superintelligence. He noted that the company is working with the Mayo Clinic to develop a healthcare foundation model, expected to be released next year. Suleyman believes that within 24 to 36 months, healthcare models will see a significant leap in capability, driven by real-world usage and rapid iteration.

This study coincides with a broader manifesto from Microsoft regarding the future of medical AI. The company aims to move beyond simple benchmarking and integrate deeply with clinical workflows. By partnering with major institutions like Beth Israel, Microsoft hopes to provide compelling evidence that these systems can be deployed at scale, potentially transforming how medical records are processed and how patients are treated in the coming years.