Microsoft AI System Outpaces Physicians in Complex Diagnosis Trials

Microsoft’s AI division has released a new diagnostic system that outperformed generalist physicians in handling complex medical cases, a development the company’s chief executive framed as a step toward medical superintelligence. The findings, published this week in the journal Science, highlight a growing trend among major tech firms to position artificial intelligence as a peer to human medical expertise.

The system, designated MAI-DxO, functions by coordinating multiple AI agents that simulate a panel of doctors. These agents interrogate patient data, request further testing, and debate potential diagnoses. In simulations using difficult cases from the New England Journal of Medicine, the system achieved an 83% accuracy rate. By comparison, primary care doctors and general internists working individually with standard reference materials were correct in 36% of instances. It is noted that such complex cases are typically handled by specialist teams in practice, who were not part of this specific comparison.

Dominic King, vice president of Health at Microsoft AI, described the results as substantial, citing the current mismatch between the demand for medical expertise and its supply. The study suggests that much of the performance improvement stems from the power of the underlying AI models, with the orchestration system providing additional gains. Microsoft tested the system using models from various providers, including OpenAI, Google, Anthropic, Meta, and DeepSeek, finding that the most significant results came from OpenAI’s GPT-5 and GPT-5.5 models.

Cost analysis revealed that while the system reduced the estimated expense per case by approximately $3,000 compared to running GPT-5 alone, the accuracy gain was modest, adding only about four percentage points. On emergency department cases, the system offered little improvement in either accuracy or cost over the standalone model. However, for complex cases, the system boosted the accuracy of the newer GPT-5.5 model from 72% to nearly 87%. The gains were most pronounced with less capable models, such as Gemini 2.5 Flash, where accuracy rose from 46.4% to nearly 69%.

Mustafa Suleyman, CEO of Microsoft AI, stated that the company is collaborating with the Mayo Clinic to train a foundation model for healthcare, with a release planned for next year. Suleyman, who previously co-founded DeepMind, believes this infrastructure will allow for significant leaps in clinical decision support. He predicted that by 2027, the industry will see a coding-style breakthrough for medical records, enabling highly accurate predictions of patient outcomes. He expects healthcare models to become an order of magnitude more capable within the next 24 to 36 months.

To validate these capabilities in real-world settings, Microsoft has partnered with Beth Israel Deaconess Medical Center in Boston. The evaluation will assess the system’s clinical reasoning across various hospital environments, including emergency and urgent care. Patients will receive standard treatment while consenting participants will also interact with the AI system. The company plans to compare the AI’s recommendations against those of human doctors. King noted that this will be one of the largest studies of its kind, providing evidence for the potential large-scale deployment of these systems.

The study coincides with a broader manifesto from Microsoft outlining the requirements for medical superintelligence, defined as surpassing contemporary human-led systems in reasoning and coordination. The company aims to move beyond benchmarking individual tasks to integrating with healthcare systems for better patient outcomes. Suleyman emphasized that Microsoft is committed to accelerating the introduction of this technology, viewing it as a key differentiator in the competitive landscape of frontier AI labs. The company already sells dictation and radiology tools and sees this diagnostic system as a natural expansion of its offerings.