Microsoft AI Diagnostic Panel Beats Doctors in Complex Case Simulation

Microsoft has unveiled a multi-agent AI system that outperforms generalist physicians in diagnosing complex medical conditions, marking a significant step toward what the company calls medical superintelligence.

The system, known as MAI-DxO or simply DxO, was tested in a simulation using difficult cases from the New England Journal of Medicine. It correctly identified the diagnosis in 83% of instances. In comparison, primary care doctors and general internists, working individually with access to reference materials, were correct only 36% of the time. While specialists or physician teams typically handle such complex cases and were not included in the direct comparison, the gap between the AI and individual generalists was substantial.

The technology operates by coordinating a panel of AI agents that mimic a medical team. These agents question the patient, order necessary tests, and debate potential diagnoses before reaching a conclusion. Microsoft tested the system using models from several major AI developers, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The highest performance was observed when the system utilized OpenAI’s GPT-5 and GPT-5.5 models.

Mustafa Suleyman, CEO of Microsoft AI, described the results as a pivotal moment for the field. He stated that the company is collaborating with the Mayo Clinic to train a dedicated foundation model for healthcare, with a release planned for next year. Suleyman believes this specialized model will allow for significant advancements, predicting that by 2027, the company will be on the verge of a breakthrough in how medical records are coded and analyzed. He envisions a future where highly accurate predictions about the trajectory of a patient's medical record are streamed to clinicians, providing foresight that he describes as invaluable.

Cost and efficiency were also measured in the study. When using GPT-5 with a basic prompt, the system achieved a 78.9% accuracy rate at an estimated cost of $8,400 per case. The DxO orchestration reduced this cost by approximately $3,000 while increasing accuracy by about four points. However, on emergency department cases, the improvement in both accuracy and cost was minimal when comparing GPT-5 with and without the DxO assist. With the newer GPT-5.5 model, released in April 2026, the results were mixed; accuracy on complex cases rose from 72% to nearly 87% at a slightly lower cost, but again, emergency department cases saw little change, with both methods scoring around 85%.

The study suggests that much of the diagnostic improvement stems from the underlying power of the AI models themselves, with the DxO orchestration layer providing additional gains. For instance, a less capable model, Gemini 2.5 Flash, improved from 46.4% accuracy on its own to nearly 69% with DxO. Despite this boost, the performance gap between weaker and stronger models remained larger than the enhancement provided by the orchestration system. Microsoft researchers noted that accuracy gains naturally diminish as models approach theoretical limits, but the system helps optimize both accuracy and cost.

Dominic King, vice president of Health at Microsoft AI, emphasized the urgency of the technology, citing a growing mismatch between the demand for and supply of medical expertise. The company is preparing for real-world evaluation through a partnership with Beth Israel Deaconess Medical Center in Boston. In this study, patients will receive standard care while those who consent will also interact with DxO. Microsoft and its partners will then compare the AI's recommendations against those of human doctors across emergency and urgent care settings.

This initiative places Microsoft in direct competition with Google and OpenAI, which have also claimed that their AI systems can match or exceed human diagnostic capabilities. Google recently published a study in The Lancet evaluating its AMIE diagnostic AI in real-world urgent care visits. King described the upcoming Microsoft study as potentially the largest of its kind, aiming to provide compelling evidence for the deployment of such systems at greater scale. The publication in Science coincides with a manifesto outlining the path to medical superintelligence, defined as surpassing human-led systems in medical reasoning across the full arc of care.

Suleyman, a co-founder of DeepMind and former head of Hark, joined Microsoft in 2024 to lead its AI unit. He expressed confidence that healthcare models will improve by an order of magnitude within the next two to three years. He likened the progress to the rapid advancements seen in coding and chat interfaces, where real-world usage creates a flywheel effect that identifies gaps and drives rapid improvement. Microsoft currently sells dictation and radiology tools to clinicians and views DxO as a critical expansion into clinical decision support, aiming to become the leading provider in this space.

The company’s strategy is to move beyond benchmarking individual tasks, such as answering medical exam questions, and instead integrate deeply with clinicians and healthcare systems to improve patient outcomes. Suleyman stated that the focus on healthcare could serve as a key differentiator for Microsoft compared to other frontier labs. He affirmed that the company is committed to accelerating the introduction of this technology, believing that the ability to provide superior diagnostic support is a capability that the global healthcare sector urgently requires.