Microsoft's AI diagnostic system correctly identified complex medical cases 83% of the time in a simulation, more than double the 36% accuracy rate of primary care doctors and general internists working alone with reference materials, according to a study published Thursday in Science. The system, called MAI-DxO, coordinates multiple AI agents that function like a medical panel, asking questions, ordering tests, and debating diagnoses. Mustafa Suleyman, CEO of Microsoft AI, described the results as progress toward "medical superintelligence."

The study tested DxO using difficult cases from the New England Journal of Medicine, though it notes these complex scenarios are typically managed by specialists or physician teams who weren't included in the comparison. Microsoft measured the system's performance across different underlying AI models, including those from OpenAI, Google, Anthropic, Meta, and DeepSeek, with the strongest results coming from OpenAI's GPT-5 and GPT-5.5. On its own with a basic prompt, GPT-5 achieved 78.9% accuracy on complex cases at roughly $8,400 per case. The DxO orchestration reduced that cost by approximately $3,000 while increasing accuracy by about four percentage points. With GPT-5.5, released in April 2026, DxO boosted accuracy from 72% to nearly 87% on complex cases at slightly lower cost, though on emergency department cases both scored around 85% with similar expenses. The orchestration system delivered its largest gains with less powerful models—Gemini 2.5 Flash jumped from 46.4% accuracy alone to nearly 69% with DxO—but the gap between weaker and stronger models remained larger than the boost DxO provided.

Dominic King, vice president of Health at Microsoft AI, told TechCrunch that "the biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise." The study's own findings suggest much of the improvement comes from more capable underlying AI models, with Microsoft's orchestration adding incremental gains. Microsoft researchers noted that accuracy improvements naturally shrink as models approach theoretical maximum performance, while a spokesperson said the orchestration helps with "the broader challenge of optimization both accuracy and cost." Suleyman said the company is partnering with Mayo Clinic to train a healthcare foundation model for release next year, which he believes will "make really enormous leaps here."

The publication positions Microsoft alongside Google and OpenAI in claiming its AI can match or exceed physician performance, and the company plans real-world testing through a partnership with Beth Israel Deaconess Medical Center in Boston. Patients who consent will receive standard care while also engaging with DxO, allowing Microsoft to compare the system's recommendations against human doctors' decisions across emergency and urgent care settings. King called it "the biggest study of its kind." Suleyman, previously a DeepMind co-founder before joining Microsoft in 2024, expects healthcare models to improve "an order of magnitude" in capability within 24 to 36 months as they enter production and iterate with real-world usage. He predicts that by 2027, Microsoft will achieve "a coding style breakthrough for medical records" that streams accurate predictions about how a patient's record might unfold, calling that foresight "gold dust" for clinicians. Microsoft already sells dictation and radiology tools to clinicians and views DxO as an expansion of those efforts, with Suleyman saying the company must have "the absolute number one state-of-the-art medical clinical decision support products over the next few years." The competitive pressure to dominate clinical AI may push every frontier lab to prioritize medical applications over other domains, potentially reshaping where the industry allocates its most advanced research talent. Whether healthcare systems can absorb these tools at the pace Microsoft envisions remains an open question that will likely determine the technology's practical impact.