Microsoft claims its AI beats doctors on complex medical cases

Mustafa Suleyman, the CEO of Microsoft AI, announced that the company's new diagnostic system has correctly identified complex medical conditions more than twice as often as generalist physicians. This achievement was detailed in a study published in Science on Thursday. Suleyman described the finding as a significant step toward what he calls medical superintelligence, a state where AI surpasses human-led health systems in reasoning and coordination.

The system, known as MAI-DxO, operates by coordinating multiple AI agents that function like a panel of doctors. These agents work together to question patients, order necessary tests, and debate potential diagnoses. In the study, the researchers used complex cases from the New England Journal of Medicine to test the system's capabilities. The AI panel was correct in 83 percent of these instances. By comparison, primary care doctors and general internists who worked individually with reference materials were correct only 36 percent of the time.

It is important to note that such complex cases are typically handled by specialists or teams of physicians in real-world settings, who were not part of this specific comparison. Dominic King, vice president of Health at Microsoft AI, told TechCrunch that the results are substantial and exciting. He highlighted that the current major challenge in healthcare is the growing mismatch between the demand for medical expertise and the available supply of skilled professionals.

With the release of these results, Microsoft joins competitors like Google and OpenAI in asserting that its AI models can match or exceed human performance in medical diagnosis. The study's data indicates that much of the improvement stems from more powerful underlying AI models, with the orchestration system adding further gains on top. Microsoft tested the system using models from various companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The best results were achieved using OpenAI's GPT-5 and GPT-5.5 models.

Cost and accuracy were both measured in the study. When using GPT-5 with a basic prompt, the system scored 78.9 percent on complex cases at an estimated cost of $8,400 per case. The DxO orchestration reduced that cost by about $3,000 while increasing accuracy by roughly four points. However, on emergency department cases, there was little improvement in either accuracy or cost when comparing the standalone model to the orchestrated version.

Results varied with the newer GPT-5.5 model, released in April 2026. For complex cases, the orchestration boosted accuracy from 72 percent to nearly 87 percent at a slightly lower cost. Again, emergency department cases showed no significant gain, with both versions scoring about 85 percent. The system provided the most benefit for less capable models. For example, Gemini 2.5 Flash improved from 46.4 percent to nearly 69 percent with the assistance of DxO, though the gap between weaker and stronger models remained larger than the boost provided by the orchestration.

Microsoft researchers noted that accuracy gains naturally shrink as models approach theoretical limits. A spokesperson added that the orchestration helps optimize both accuracy and cost. Suleyman revealed that the company is collaborating with the Mayo Clinic to train a foundation model for healthcare, with a release planned for next year. He believes this model will enable enormous leaps in the field. He predicts that by 2027, the company will be on the verge of a breakthrough in handling medical records, allowing for highly accurate predictions about how a record might unfold, which he described as gold dust for clinicians.

Suleyman, who previously co-founded DeepMind and joined Microsoft in 2024, sees this as the next phase of his career in medical AI. He expects healthcare models to become an order of magnitude better in capability within the next 24 to 36 months. He explained that progress accelerates once a flywheel is established, where models in production are iterated upon using real-world data to identify and fix gaps. This pattern has been seen in coding and chat applications and is expected to drive progress in medical records and clinical decision support next year.

To validate these findings, Microsoft has partnered with Beth Israel Deaconess Medical Center in Boston to evaluate the system in real-world settings. This study will compare DxO recommendations against human doctors across different hospital environments, including emergency and urgent care. King called this the biggest study of its kind, noting that it will provide compelling evidence for deploying these systems at a greater scale in the future.

The study coincides with a manifesto outlining the path to medical superintelligence. Microsoft aims to move beyond benchmarking individual tasks to integrating with clinicians and healthcare systems for better outcomes. Suleyman believes this focus could differentiate Microsoft from other frontier labs. The company already sells dictation and radiology tools to clinicians and views DxO as an expansion of those efforts. He stated that the company must have the number one state-of-the-art medical clinical decision support product over the next few years.