Medical Superintelligence: Microsoft AI Outperforms Physicians in Diagnostics

Microsoft is advancing its ambition in healthcare with a new study indicating its artificial intelligence diagnostic system can significantly outperform human doctors on complex cases. Published on Thursday, October 22, 2026, in the journal Science, the research details the performance of a system called MAI-DxO, which Mustafa Suleyman, CEO of Microsoft AI, described as a step toward 'medical superintelligence.' The development signals a new phase in the competition among technology firms to apply advanced AI to medical diagnosis and decision-making.

The MAI-DxO system functions by coordinating multiple AI agents that simulate a panel of doctors. This group of agents interacts with a patient case, requests tests, and debates potential diagnoses to arrive at a conclusion. In a simulation based on complex cases from the New England Journal of Medicine, the system achieved a correct diagnosis 83% of the time. This compared to a 36% accuracy rate for primary care physicians and general internists working individually with reference materials. The source material notes that cases of this complexity are typically handled by specialists or teams of physicians, who were not part of this particular comparison.

Microsoft's publication of these results places it alongside Google and OpenAI, both of which have also claimed their AI models can match or surpass human performance in medical tasks. Dominic King, vice president of Health at Microsoft AI, commented on the study's implications, stating, 'There is something very substantial and exciting here.' He added, 'The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise.' The study's own data indicates that a portion of the accuracy gains are attributable to more powerful underlying AI models, with Microsoft's orchestration system providing additional improvements.

The system's performance varied depending on the underlying AI model and the type of case. For example, Microsoft tested MAI-DxO with models from several companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek. Its strongest results were achieved using OpenAI's GPT-5 and GPT-5.5. On complex cases, the use of DxO with GPT-5.5 increased accuracy from 72% to nearly 87% at a slightly lower cost. However, on emergency department cases, DxO did not provide a significant boost in accuracy or cost savings when paired with either GPT-5 or GPT-5.5. Microsoft's researchers said that accuracy gains naturally diminish as models approach theoretical maximums, and that 'The DxO orchestration helps with the broader challenge of optimization both accuracy and cost.'

Looking ahead, Microsoft is pursuing real-world evaluations of its technology. The company is working with the Mayo Clinic to develop a healthcare-specific foundation model planned for release next year. Suleyman believes this work will lead to major advancements, stating, 'I really believe that in 2027, we will be on the verge of a coding style breakthrough for medical records.' He anticipates the ability to generate highly accurate predictions about the progression of a patient's medical history. Furthermore, Microsoft has established a partnership with Beth Israel Deaconess Medical Center in Boston to evaluate the clinical reasoning capabilities of DxO across different hospital settings.

Through this partnership, patients will receive standard care as determined by human clinicians, while consenting participants will also engage with the DxO system. Microsoft and its partners will then compare the AI's recommendations against the decisions made by human doctors. King described the potential scope of this evaluation, stating, 'This technology is ready for real-world evaluations, and you can imagine the complexity and diversity of issues that come from a Boston emergency department on a Friday evening.' He added, 'We think this will be the biggest study of its kind.. and will give us very compelling evidence by the end of the study period around the ability for these systems to be deployed at greater scale in the future.'