Microsoft Claims Its AI Panel Beats Generalists on Tough Medical Cases

For decades, the medical field has treated artificial intelligence as a distant possibility, but a new report published this week suggests that capability is arriving faster than many experts predicted. The shift is not just about adding a chatbot to a clinic; it is about restructuring how difficult diagnoses are made.

According to a study released in the journal *Science* on Thursday, Microsoft’s latest diagnostic tool, known as MAI-DxO or DxO, demonstrated a marked advantage over human generalists. In a simulation using complex cases from the *New England Journal of Medicine*, the system reached a correct diagnosis 83% of the time. By comparison, primary care doctors and general internists, who worked individually with reference materials, were only correct 36% of the time. It is important to note that these specific cases are typically handled by specialists or teams of physicians, who were not included in the direct comparison, making the gap against generalists particularly stark.

The system does not function as a single brain. Instead, it coordinates a panel of AI agents that behave like a medical board. These agents question the patient, order necessary tests, and debate the findings to reach a consensus. Mustafa Suleyman, the CEO of Microsoft AI, described this achievement as a significant step toward what the company is calling “medical superintelligence.”

While the accuracy numbers are impressive, the study reveals that much of the improvement comes from the power of the underlying models rather than the orchestration system itself. Microsoft tested DxO using models from several competitors, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The best results were achieved with OpenAI’s GPT-5 and the newer GPT-5.5, released in April 2026. When using GPT-5.5, DxO boosted accuracy on complex cases from 72% to nearly 87% at a slightly lower cost. However, this boost was not consistent across all scenarios. In emergency department cases, the system showed little to no improvement in either accuracy or cost efficiency compared to running the models without the orchestration layer.

Cost was a key metric in the evaluation. On its own, with a basic prompt, GPT-5 scored 78.9% on complex cases but carried an estimated cost of roughly $8,400 per case. DxO reduced that cost by about $3,000 while raising accuracy by only four points. The system proved most beneficial when paired with less capable models. For instance, Google’s Gemini 2.5 Flash scored 46.4% on its own but jumped to nearly 69% with the help of DxO. Even so, the performance gap between the weaker and stronger base models was larger than the boost provided by the orchestration system.

Dominic King, vice president of Health at Microsoft AI, emphasized the practical implications of these results. He noted that the primary challenge in healthcare is the growing mismatch between the demand for medical expertise and the available supply. “There is something very substantial and exciting here,” King said. “The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise.”

Microsoft is not the only company making these claims. The publication of these results places the tech giant alongside Google and OpenAI, both of which have recently asserted that their AI can match or outperform human doctors. Google, for example, published a clinical study alongside Beth Israel Deaconess Medical Center in *The Lancet* this month, evaluating its own conversational diagnostic AI in real-world urgent care visits.

To validate its own technology, Microsoft is partnering with Beth Israel Deaconess Medical Center in Boston for a large-scale real-world evaluation. The study will compare DxO’s recommendations against those of human doctors in various hospital settings, including emergency and urgent care environments. Patients will receive standard treatment, while those who consent will also interact with the AI system in parallel. King believes this will be the largest study of its kind, providing compelling evidence on the system's ability to scale.

Suleyman, who previously co-founded DeepMind and led its health initiatives before joining Microsoft in 2024, is pushing for rapid progress. He stated that the company is working with the Mayo Clinic to train a specific healthcare foundation model, expected to release next year. He believes this will allow for “really enormous leaps” in medical reasoning. “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.”

The company has released a manifesto defining “medical superintelligence” as the capacity to surpass the strongest human-led health systems in reasoning and coordination. Microsoft aims to move beyond simple benchmarking tasks and integrate deeply with clinical workflows. Suleyman emphasized that the goal is to become the leader in clinical decision support, a sector where the company already sells dictation and radiology tools. “We have to have the absolute number one state-of-the-art medical clinical decision support products over the next few years,” he said, noting that healthcare models should become an order of magnitude better in capability within the next two to three years.