Microsoft's AI System Beats General Doctors on Complex Cases

83 percent. That is the accuracy rate achieved by Microsoft’s new diagnostic AI, MAI-DxO, when simulating complex medical cases. The figure stands in stark contrast to the 36 percent success rate recorded by generalist physicians working individually with reference materials. These results were detailed in a study published this week in the journal Science.

The system operates like a virtual panel of specialists. Multiple AI agents collaborate to interview a patient, order necessary tests, and debate potential diagnoses. While the study focused on complex cases typically handled by specialists or teams, the comparison highlights a significant gap between current AI capabilities and individual generalist performance. Dominic King, vice president of Health at Microsoft AI, described the findings as substantial. He noted that the growing mismatch between medical demand and supply is a primary challenge in healthcare delivery.

Performance varied depending on the underlying model used. Microsoft tested the orchestration system with models from OpenAI, Google, Anthropic, Meta, and DeepSeek. The highest scores came when paired with OpenAI’s GPT-5 and GPT-5.5. For complex cases, GPT-5 alone scored 78.9 percent at a cost of roughly $8,400 per case. Adding the MAI-DxO layer raised accuracy by about four points while reducing the estimated cost by approximately $3,000.

The newer GPT-5.5 model showed different results. On complex cases, the orchestration system boosted accuracy from 72 percent to nearly 87 percent at a slightly lower cost. However, for emergency department scenarios, the advantage was minimal. Both the standalone model and the assisted system scored around 85 percent at similar costs. Researchers noted that accuracy gains naturally diminish as models approach their theoretical limits. The system provided the largest boost for weaker models, such as Gemini 2.5 Flash, which improved from 46.4 percent to nearly 69 percent with the assistance.

Mustafa Suleyman, CEO of Microsoft AI, views this work as a foundation for achieving what he calls medical superintelligence. This concept is defined as surpassing the strongest human-led health systems in reasoning and coordination. Suleyman stated that the company is collaborating with the Mayo Clinic to train a foundation model for healthcare, with a release planned for next year. He believes this infrastructure will allow for enormous leaps in clinical decision support.

The company is also preparing for real-world validation. Through a partnership with Beth Israel Deaconess Medical Center in Boston, Microsoft will evaluate the system across various hospital settings. Patients will receive standard care, while those who consent will also interact with the AI. The goal is to compare AI recommendations against human decisions in diverse environments, including urgent care and emergency rooms. King called this the largest study of its kind, aiming to provide compelling evidence for broader deployment.

Suleyman, who previously co-founded DeepMind and joined Microsoft in 2024, expects rapid progress in the coming years. He predicted that healthcare models will become an order of magnitude more capable within the next 24 to 36 months. He cited the development of medical records as a key area, envisioning a coding-style breakthrough by 2027 that allows for streaming accurate predictions of patient outcomes. This foresight, he argued, is invaluable to clinicians. Microsoft aims to make these clinical decision support tools a core differentiator in the competitive AI landscape.

The company already offers dictation and radiology tools to medical professionals. The new diagnostic system represents an expansion of those efforts. Suleyman emphasized the need for state-of-the-art products in the coming years. He expressed confidence that the combination of strong base models and specialized orchestration will drive the necessary advancements in medical AI.