A new study published in the journal Science has found that Microsoft's AI-powered diagnostic system outperformed generalist physicians on complex medical cases by more than twofold in a controlled simulation. The findings have prompted Mustafa Suleyman, chief executive of Microsoft AI, to describe the outcome as a meaningful step toward what he calls "medical superintelligence."

The system at the center of the research is called MAI-DxO, commonly shortened to DxO. Rather than functioning as a single AI model, it operates by coordinating multiple AI agents that collectively simulate the deliberative process of a physician panel. These agents interact with patient information, request diagnostic tests, and debate potential diagnoses before arriving at a conclusion.

When tested against a set of challenging cases drawn from the New England Journal of Medicine, DxO arrived at the correct diagnosis 83% of the time. By comparison, primary care physicians and general internists working individually - and with access to reference materials - achieved a correct diagnosis rate of just 36%. It is worth noting, however, that cases of this complexity are typically referred to specialists or multidisciplinary teams, neither of whom were included in the comparison group.

Dominic King, vice president of Health at Microsoft AI, highlighted the broader healthcare context surrounding the findings. "The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise," he said. King described the results as "very substantial and exciting."

Joining a Competitive Field

With this publication, Microsoft enters a growing list of technology companies asserting that their AI systems can match or surpass the diagnostic capabilities of human doctors. Both Google and OpenAI have previously made similar claims based on their own research and benchmarking efforts.

The study's internal data also reveals an important nuance: a significant portion of the performance gains observed appear to stem from improvements in the underlying AI models themselves, with DxO's orchestration layer contributing additional but incremental gains on top of those baseline improvements.

Microsoft tested DxO using models from a range of AI developers, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The strongest results, however, were achieved when DxO was paired with OpenAI's GPT-5 and GPT-5.5 models.

Performance and Cost Breakdown

Microsoft evaluated DxO not only on diagnostic accuracy but also on the estimated cost of running the system through each case. When GPT-5 operated independently with a basic prompt, it achieved a 78.9% accuracy rate on complex cases at an estimated cost of approximately $8,400 per case. Deploying DxO alongside GPT-5 reduced that cost by roughly $3,000, while improving accuracy by only around four percentage points. In emergency department scenarios, DxO produced little measurable improvement in either accuracy or cost efficiency compared to GPT-5 working alone.

Results with GPT-5.5, a newer model released in April 2026, were more varied. On complex diagnostic cases, DxO raised GPT-5.5's accuracy from 72% to nearly 87% at a slightly lower cost - a meaningful improvement. However, in emergency department settings, the gains once again failed to materialize. Both the standalone and DxO-assisted configurations of GPT-5.5 scored around 85% at comparable costs.

The pattern that emerges from the data is that DxO delivers its most significant benefits when applied to complex cases and when paired with less capable underlying models. For instance, Google's Gemini 2.5 Flash scored 46.4% on its own but climbed to nearly 69% when paired with DxO. Even so, the performance gap between weaker and stronger base models was considerably larger than the boost DxO alone could provide.

Microsoft's research team acknowledged that accuracy gains tend to diminish naturally as models approach theoretical performance ceilings. A company spokesperson noted that "the DxO orchestration helps with the broader challenge of optimizing both accuracy and cost."

Plans for a Healthcare Foundation Model

Suleyman, who oversees Microsoft's broader effort to develop its own frontier AI systems, confirmed that the company is working with the Mayo Clinic on a foundation model specifically designed for healthcare applications, with a planned release scheduled for next year. He expressed confidence that this model will enable significant advances, describing DxO as part of the core infrastructure Microsoft is assembling to drive real breakthroughs in medicine.

"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's focus on applying AI to medicine spans his entire professional career. He was previously a co-founder of DeepMind, the AI research lab that was acquired by Google in 2014. The following year, Google DeepMind acquired Hark, a clinical task management application founded by Dominic King, and the two subsequently launched DeepMind Health as a joint initiative. Suleyman joined Microsoft in 2024 to lead its AI division, with King following him to the company a few months later.

"We're just true believers," Suleyman said of their shared commitment to the mission.

While acknowledging that the application of generative AI in healthcare remains at an early stage, Suleyman said he anticipates healthcare-focused AI models improving by "an order of magnitude" in capability over the next 24 to 36 months.

"As we've seen multiple times in the last five or six years, the progress happens very quickly once you have a flywheel. A flywheel means we have a model in production which we are iterating with real-world usage in order to identify gaps in the model and then patch them and improve them. That's what's happened in coding. It's what happened in chat before that. And I think next year it's going to start to happen for medical records and for clinical decision support."

Real-World Trials at Beth Israel

Through a newly announced partnership with Beth Israel Deaconess Medical Center in Boston, Microsoft will move DxO beyond simulation-based benchmarks and into real clinical environments. The collaboration is designed to evaluate DxO's clinical reasoning capabilities across a variety of hospital settings.

"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," King said.

King outlined how the trial will be structured. Patients will receive standard care throughout the study, including triage, specialist consultations, treatment, investigations, and discharge decisions. In parallel, those who provide consent will also interact with DxO. Microsoft and its clinical partners will then compare the system's recommendations against those made by the treating physicians.

The announcement comes shortly after Google published its own clinical study alongside Beth Israel Deaconess in The Lancet, evaluating its AMIE conversational diagnostic AI during real-world urgent care visits, with human physicians monitoring the AI's interactions in real time.

King described Microsoft's forthcoming study as potentially the most comprehensive of its kind. "We think this will be the biggest study of its kind," he said. "It will be across not just emergency settings but urgent care environments, and will give us very compelling evidence by the end of the study period around the ability for these systems to be deployed at greater scale in the future."

Defining Medical Superintelligence

Microsoft's publication in Science is accompanied by a broader strategic document that functions as a framework for what the company believes it will take to achieve what it terms medical superintelligence. The document defines this concept as "the capacity to surpass the strongest contemporary human-led health systems in medical reasoning and coordination across the full arc of care."

The company's stated ambition is to move beyond evaluating AI systems on isolated tasks - such as passing medical licensing exams or generating individual diagnoses - and toward deeper integration with clinicians and healthcare systems aimed at producing measurably better patient outcomes.

"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. He suggested that this level of commitment to healthcare AI could serve as a meaningful differentiator for Microsoft relative to other frontier AI developers.

Microsoft already has an established presence in clinical settings, offering dictation tools and radiology assistance products to healthcare providers. The company views DxO as a natural extension of those existing offerings into the domain of clinical decision support.

"We have to have the absolute number one state-of-the-art medical clinical decision support products over the next few years," Suleyman said, underscoring the priority the company is placing on this area as competition among AI developers in the healthcare sector continues to intensify.