Microsoft’s Mustafa Suleyman says its AI outperforms doctors in difficult cases as it pursues ‘medical superintelligence’

Microsoft’s AI diagnostic system correctly diagnosed complex medical cases in a simulation more than twice as often as general practitioners, according to a study published Thursday in Science. Mustafa Suleyman, CEO of Microsoft AI, called the result a step toward “medical superintelligence.”

The system, called MAI-DxO (DxO for short), coordinates several AI agents that act like a panel of doctors. They interview a patient, order tests, and debate a diagnosis.

In a simulation using complex cases from the New England Journal of Medicine, this was correct 83% of the time. Primary care physicians and general internists, working individually with reference material, were correct in 36% of cases. A note: such complex cases are generally treated by specialists or teams of doctors, who were not part of the comparison.

“There’s something very substantial and exciting here,” Dominic King, vice president of health at Microsoft AI, told TechCrunch. “Today’s greatest challenge in healthcare delivery is the growing mismatch between demand and supply of medical expertise. »

By publishing these results, Microsoft joins Google and OpenAI in claiming that its AI can match or surpass doctors.

The study’s own data suggests that much of the improvement comes from more powerful underlying AI models, with Microsoft’s orchestration system adding further gains.

Microsoft tested MAI-DxO with models from several companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek, but its best results came with OpenAI’s GPT-5 and GPT-5.5.

Microsoft measured DxO based on cost as well as accuracy, that is, the estimated cost of running the AI ​​in each case. On its own, with a short, basic prompt, GPT-5 scored 78.9% on complex cases, at around $8,400 per case. DxO reduced that cost by about $3,000 but only increased accuracy by about four points. In the cases of emergency services, there has been little improvement in accuracy or cost improvement between GPT-5 alone and with the help of DxO.

With GPT-5.5, a newer model released in April 2026, the results are mixed. On complex cases, DxO increased the accuracy score of GPT-5.5 from 72% to almost 87% for a slightly lower cost. However, once again, the gains were not evident in emergency department cases. Both (with and without DxO) scored around 85% for a similar cost.

DxO is more useful in complex cases and with less efficient AI models. Gemini 2.5 Flash, for example, scores 46.4% on its own but almost 69% with DxO. Despite this, the gap between the weakest and strongest models was larger than the boost provided by DxO.

Microsoft researchers said accuracy gains naturally decline as models approach the theoretical maximum. “DxO orchestration helps address the broader challenge of optimizing accuracy and cost,” a spokesperson added.

Suleyman, who is leading Microsoft’s efforts to create its own cutting-edge AI, told TechCrunch that the company is working with the Mayo Clinic to train a basic model for healthcare, scheduled for release next year. He believes this model will allow the company to “make really huge strides here” and says DxO is part of the foundational infrastructure that Microsoft is building to make real breakthroughs in healthcare.

“I truly believe that in 2027 we will be on the verge of revolutionizing the style of coding medical records,” Suleyman said. “I think we’re going to be able to put out very accurate predictions about how a medical record might turn out, and obviously that kind of foresight for a clinician is gold dust.”

Suleyman has spent his career trying to use AI to solve medical problems. He was previously a co-founder of DeepMind, which was acquired by Google 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 followed a few months later.

“We are just true believers,” Suleyman said.

He said that while the use of generative AI in healthcare is still in its early stages, he expects healthcare models to become “an order of magnitude greater in capabilities over the next 24 to 36 months.”

“As we’ve seen repeatedly over the last five or six years, progress happens very quickly once you have a flywheel,” Suleyman said. “A flywheel means we have a model in production that we iterate with real use in order to identify gaps in the model and then fix them and improve them. That’s what happened in coding. That’s what happened in chat before that. And I think next year that’s going to start happening for medical records and for clinical decision support.”

Through a new partnership with Beth Israel Deaconess Medical Center in Boston, Microsoft will evaluate DxO’s clinical reasoning capabilities in different hospital settings.

“This technology is ready for real-world assessments, and you can imagine the complexity and diversity of problems that arise in a Boston emergency department on a Friday evening,” King said.

He explained how the study works: Patients will receive standard treatment, including triage, time with a specialist, treatments, investigations and a decision about whether it is safe to be discharged from the hospital. At the same time, those who consent will also commit to DxO. Microsoft and its partners will compare DxO’s recommendations to those of human doctors.

Google this month published a clinical study alongside Beth Israel in The Lancet to evaluate Google’s AMIE conversational diagnostic AI during real-world emergency care visits, while human doctors monitored the interactions in real time.

“We think this will be the largest study of its kind,” King said. “This will address not only emergency settings but also urgent care environments, and will provide us with very compelling evidence by the end of the study period on the ability of these systems to be deployed on a larger scale in the future. »

Microsoft’s study in Science coincides with a manifesto of sorts on what it will take to build medical superintelligence, defined as “the ability to outperform the most powerful contemporary human-led healthcare systems in medical reasoning and coordination across the spectrum of care.” Microsoft hopes to go beyond benchmarking individual tasks — like answering medical exam questions and generating diagnoses — by integrating with clinicians and health systems for better outcomes.

“This is the kind of technology that everyone wants, and that’s why we are committed to doing everything we can to accelerate its introduction into the world,” Suleyman said, adding that this focus could be a differentiator for Microsoft compared to other pioneering labs.

Microsoft already sells dictation and radiology tools to clinicians and sees DxO as an extension of those efforts.

“We need to have the number one cutting-edge medical clinical decision support products in the coming years,” Suleyman said.