Microsoft’s AI diagnostic system correctly diagnosed complex medical cases more than twice as often as generalist physicians in a simulation, according to a study published Thursday, October 22, in the journal Science. The system, called MAI-DxO, scored 83% on hard cases drawn from the New England Journal of Medicine; primary care doctors and general internists working individually with reference materials scored 36%. Mustafa Suleyman, CEO of Microsoft AI, called the result a step toward “medical superintelligence.” The headline number is real, but TechCrunch’s reporting highlights several caveats about what it does and does not show.

How MAI-DxO works

MAI-DxO, often shortened to DxO, is not a single model. It coordinates several AI agents that act like a panel of doctors: they question a simulated patient, order tests and debate a diagnosis. In other words, it is an orchestration layer that sits on top of existing language models. Microsoft tested it with models from several companies, including OpenAI, Google, Anthropic, Meta and DeepSeek. The best results came with OpenAI’s GPT-5 and GPT-5.5.

The numbers

The study used complex cases from the New England Journal of Medicine, known as CPC cases. Key results as reported by TechCrunch:

  • DxO overall: correct in 83% of the complex simulated cases.
  • Generalist physicians: 36% correct, working individually with reference materials.
  • GPT-5 alone: 78.9% on complex cases at about $8,400 per case in estimated running cost. DxO cut that cost by about $3,000 but raised accuracy by only about four points.
  • GPT-5.5 (released in April 2026): accuracy rose from 72% to nearly 87% with DxO, at a slightly lower cost.
  • Gemini 2.5 Flash: 46.4% alone, nearly 69% with DxO.
  • Emergency department cases: little or no improvement from DxO. With GPT-5.5, both setups scored about 85% at similar cost.

Caveats that matter

First, the comparison group is narrow. TechCrunch notes that cases this complex are usually handled by specialists or physician teams, who were not part of the comparison. Beating individual generalists on cases they would normally refer onward says less than beating a specialist panel would.

Second, much of the improvement appears to come from stronger underlying models rather than from the orchestration. According to the study’s own data, as summarized by TechCrunch, the gap between weaker and stronger models was larger than the boost DxO added. The orchestration helps most on complex cases and with less capable models, and its gains shrink as models approach the ceiling.

Third, this is a simulation. Questioning a simulated patient from a case write-up is not the same as handling an undifferentiated patient in a clinic, with incomplete records, time pressure and legal responsibility. The result has been published in a peer-reviewed journal, but it does not by itself show safety or benefit in clinical practice.

Context: a crowded claim

TechCrunch notes that in publishing these results Microsoft joins Google and OpenAI in claiming its AI can match or outperform doctors. Each company uses its own test sets and conditions, so the claims are hard to compare directly. Microsoft’s own cost reporting is unusual and useful: it frames diagnosis as an accuracy-versus-cost trade-off, since ordering more tests increases both information and expense.

What Microsoft says it will do next

Suleyman told TechCrunch that Microsoft is working with the Mayo Clinic to train a foundation model for healthcare, with a release planned for next year. Dominic King, vice president of Health at Microsoft AI, framed the motivation as a growing mismatch between demand for and supply of medical expertise. For background on where AI is already used in care, see how AI is used in healthcare and our look at risks and regulation.

What to watch next

  • Prospective studies that test MAI-DxO or similar systems with real patients and clinicians.
  • Comparisons that include specialists or physician teams, not only individual generalists.
  • Regulatory classification of diagnostic orchestration systems.
  • Details of the Mayo Clinic healthcare foundation model planned for next year.
  • Independent replication with other case sets, including emergency medicine.

Key takeaways

  • MAI-DxO scored 83% on complex NEJM cases; generalist physicians scored 36% in the same simulation.
  • The system coordinates multiple AI agents that question a patient, order tests and debate a diagnosis.
  • Specialists and physician teams, who usually handle such cases, were not compared.
  • Much of the gain comes from stronger base models; DxO adds more for weaker models and complex cases.
  • It was a simulation; no clinical benefit or safety has been shown in practice.

Frequently asked questions

Does this mean AI is better than doctors?

Not in general. It outperformed individual generalists on very hard simulated cases; specialists were not tested, and real-world care involves much more than reaching a diagnosis.

What is MAI-DxO?

It is Microsoft’s orchestration system that makes several AI agents work like a panel of doctors, using models from providers such as OpenAI and Google.

Is it available to patients?

The reporting does not describe a patient-facing release. Microsoft says it plans a healthcare foundation model with the Mayo Clinic for next year.