Microsoft claims AI panel beats doctors on complex diagnoses

Microsoft’s artificial intelligence system has demonstrated the ability to outperform human physicians on difficult medical cases, a development that intensifies the competition among tech giants to claim the future of clinical decision-making. In a study published Thursday in Science, Microsoft’s diagnostic tool, MAI-DxO, correctly identified complex conditions in a simulation more than twice as often as generalist doctors. Mustafa Suleyman, the CEO of Microsoft AI, described the achievement as a critical step toward what the company is calling medical superintelligence, a state where AI surpasses human-led health systems in reasoning and coordination.

The timeline of this development reflects a rapid escalation in AI’s healthcare ambitions. Microsoft’s system operates as a coordinated panel of AI agents that mimic a medical board, questioning patients, ordering tests, and debating diagnoses. In the study, which used 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. However, researchers noted that such complex cases are typically handled by specialists or teams in real-world settings, a group that was not included in the direct comparison.

The results suggest that much of the performance boost comes from more powerful underlying models rather than the orchestration system alone. Microsoft tested the system with models from various firms, including OpenAI, Google, and Anthropic, but found the best results using OpenAI’s GPT-5 and GPT-5.5. While the multi-agent system improved accuracy and reduced costs for complex cases, the gains were minimal for emergency department scenarios. For instance, with GPT-5.5, the system raised accuracy from 72% to nearly 87% on complex cases, but both the AI and human doctors scored around 85% on emergency cases, showing little difference in performance or cost efficiency in that specific context.

Looking ahead, Microsoft is moving from simulation to real-world application. The company is partnering with the Mayo Clinic to train a healthcare-specific foundation model, with a release planned for next year. Suleyman predicts that by 2027, the technology will enable streaming, highly accurate predictions of how medical records might unfold, offering clinicians significant foresight. He anticipates that healthcare models will become an order of magnitude more capable within the next two to three years, driven by a flywheel of real-world usage and iteration.

To validate these claims, Microsoft will evaluate the system in actual hospital settings through a new partnership with Beth Israel Deaconess Medical Center in Boston. This study will compare the AI’s recommendations against those of human doctors across various environments, including urgent care and emergency rooms. Patients will receive standard treatment while consenting to parallel AI analysis. This follows a similar study published by Google earlier this month, which tested its conversational diagnostic AI in urgent care visits. Microsoft aims to use these real-world evaluations to prove that its technology is ready for large-scale deployment, positioning itself as a leader in clinical decision support.

The study coincides with a broader manifesto from Microsoft outlining the path to medical superintelligence. The company seeks to move beyond simple benchmarking, such as answering exam questions, toward deep integration with clinicians and healthcare systems. Suleyman believes this focus could differentiate Microsoft from other frontier labs, as the company already sells dictation and radiology tools to the medical community. He remains confident that the rapid progress seen in other AI fields will soon be replicated in clinical decision support, stating that the company is committed to accelerating the introduction of this technology to the world.