Microsoft's MAI-DxO tops doctors 83% to 36% in Science study

Microsoft's study joins similar claims from Google and OpenAI that their AI can match or outperform doctors, but the details reveal a more nuanced picture. The system was tested with models from several companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek, and its best results came with OpenAI's GPT-5 and GPT-5.5. Yet the study's own data indicates that the gap between weaker and stronger models was larger than the boost DxO provided, suggesting the orchestration layer adds only incremental value on top of model capability.

Mustafa Suleyman, Microsoft AI's CEO, called the result a step toward "medical superintelligence" and said the company is working with Mayo Clinic to train a foundation model for healthcare, planned for release next year. He expects that model to enable "really enormous leaps" and predicted that by 2027 there will be a "coding style breakthrough for medical records," with the ability to stream highly accurate predictions about how a medical record might unfold.

The study coincides with a Microsoft manifesto defining medical superintelligence as "the capacity to surpass the strongest contemporary human-led health systems in medical reasoning and coordination across the full arc of care." Microsoft already sells dictation and radiology tools to clinicians and sees DxO as an expansion of those efforts. Through a new partnership with Beth Israel Deaconess Medical Center in Boston, Microsoft will evaluate DxO's clinical reasoning across different hospital settings, which King described as potentially "the biggest study of its kind."