Microsoft’s MAI-DxO system correctly diagnosed complex medical cases more than twice as often as generalist physicians, according to a study published Thursday in Science. It scored 83% on complex cases drawn from the New England Journal of Medicine. Primary care doctors and general internists working alone, even with reference materials, managed 36%.
That gap matters. But the study itself raises an asterisk: cases this complex usually go to specialists or physician teams, and those teams weren’t included in the comparison. The benchmark pits a machine against a single generalist, not the collaborative care patients with difficult cases typically receive. Patients and clinicians deserve to know exactly what the study measured before anyone treats these numbers as a verdict on who should make diagnoses.
How the System Works
MAI-DxO, also called DxO, coordinates several AI agents that act like a panel of doctors. They question patients, order tests and debate diagnoses with one another. Microsoft tested the system with models from OpenAI, Google, Anthropic, Meta and DeepSeek; its best results came with OpenAI’s GPT-5 and GPT-5.5. Study data indicated that more powerful underlying models accounted for much of the improvement, while Microsoft’s orchestration system added further gains.
Microsoft has joined Google and OpenAI in saying its AI can match or outperform doctors. The field is crowded now, and the people most affected by errors—patients in ordinary clinics and emergency rooms—have had little say in how these tools are built or tested.
Cost and Accuracy Trade-Offs
Microsoft measured estimated operating costs alongside accuracy. With a short, basic prompt, GPT-5 scored 78.9% on complex cases at about $8,400 per case. DxO cut the cost by about $3,000 and lifted accuracy by about four percentage points. On emergency department cases, GPT-5 showed little improvement in accuracy or cost, with or without DxO.
GPT-5.5, released in April 2026, scored 72% on complex cases on its own. With DxO, its accuracy climbed to nearly 87% at a slightly lower cost. On emergency department cases, both versions scored about 85% at similar cost. Gemini 2.5 Flash scored 46.4% alone and nearly 69% with DxO.
DxO helped most on complex cases and with less capable models. Still, the gap between weaker and stronger models exceeded the boost DxO provided. Microsoft researchers said accuracy gains naturally shrink as models approach the theoretical maximum. A spokesperson said, “The DxO orchestration helps with the broader challenge of optimization both accuracy and cost.”
Who Gets the Benefit
Dominic King, vice president of Health at Microsoft AI, told TechCrunch: “There is something very substantial and exciting here. The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise.” That shortage is real. It hits hardest among patients who already wait longest. Whether cheaper diagnostic tools reach rural clinics and safety-net hospitals, or mainly well-funded systems that can afford the license, is a question the coverage doesn’t answer.
Mustafa Suleyman, Microsoft AI’s CEO, called the result a step toward “medical superintelligence.” He said the company is working with the Mayo Clinic to train a healthcare foundation model, with release planned for next year.
“I really believe that in 2027, we will be on the verge of a coding style breakthrough for medical records,” Suleyman said. “I think we’re going to be able to stream highly accurate predictions about how a medical record might unfold, and obviously that kind of foresight for a clinician is gold dust.”
Suleyman co-founded DeepMind, which Google acquired in 2014. The following year, Google DeepMind acquired King’s startup Hark, a clinical task management app, and the two launched DeepMind Health together. Suleyman joined Microsoft to lead its AI unit in 2024, and King joined a few months later. “We’re just true believers,” Suleyman said.
Testing in the Real World
Microsoft plans to evaluate DxO’s clinical reasoning across hospital settings through a new partnership with Beth Israel Deaconess Medical Center in Boston. King said: “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.”
Patients in the study will receive standard treatment, including triage, time with a specialist, treatments, investigations and a decision on whether it’s safe to discharge. Patients who consent will also engage with DxO, and Microsoft and its partners will compare the system’s recommendations with those of human doctors. King said the study would include emergency and urgent-care settings and provide evidence by the end of the study period on whether the systems could be deployed at greater scale.
That consent step is where public oversight must do its work. Patients should know when an algorithm is part of their care, and independent reviewers—not the company that built the product—should judge the outcomes.
Google published a clinical study with Beth Israel in The Lancet this month, evaluating its AMIE conversational diagnostic AI during real-world urgent-care visits, with human physicians monitoring interactions in real time. Microsoft’s Science study coincided with a manifesto describing 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.”
“This is the kind of technology that everybody in the world wants, and that’s why we’re committed to doing everything we can to accelerate its introduction to the world,” Suleyman said. Microsoft already sells dictation and radiology tools to clinicians and sees DxO as an expansion of those products. “We have to have the absolute number one state-of-the-art medical clinical decision support products over the next few years,” Suleyman said.
Why This Matters:
The promise is real. So are the stakes. A system that helps a clinician reach a correct diagnosis more often could spare patients the delays and misreads that hit people with the fewest resources hardest. But the figures released so far compare an AI panel with a lone generalist, not the specialist teams who handle these cases in practice, and the company selling the product reports the cost savings. Patients in an emergency department on a Friday evening will live with the results of this rollout. Whether Beth Israel and other trials get independent review, patients learn when an algorithm shapes their care, and benefits reach public hospitals—not just wealthy ones—will determine whether this becomes a tool for broader access or another technology that widens the gap between those who can afford good care and those who can’t.
- microsoft
- mai-dxo
- medical diagnosis
- artificial intelligence
- mustafa suleyman
- healthcare
- beth israel deaconess medical center
- clinical study