Microsoft AI Outperforms Human Doctors on Complex Cases

Microsoft has announced that its new AI diagnostic system, MAI-DxO, correctly identified complex medical conditions in simulated scenarios at a rate more than double that of generalist physicians. The results, published in the journal Science on Thursday, mark a significant milestone for the company. Mustafa Suleyman, CEO of Microsoft AI, described the achievement as a critical step toward achieving what he calls medical superintelligence.

The system, known as DxO for short, operates by coordinating multiple AI agents that function like a panel of specialists. These agents interrogate patient data, order virtual tests, and debate potential diagnoses. In simulations using difficult cases from the New England Journal of Medicine, the system achieved an accuracy rate of 83%. By comparison, primary care doctors and general internists working individually with reference materials were correct only 36% of the time. It is worth noting that such complex cases are typically handled by specialist teams in real clinical settings, who were not included in this specific comparison.

The path to this announcement began with the integration of advanced language models into a coordinated framework. Microsoft tested the system using models from OpenAI, Google, Anthropic, Meta, and DeepSeek. The highest performance was recorded using OpenAI’s GPT-5 and GPT-5.5 models. The study indicates that much of the accuracy improvement stems from the power of the underlying models themselves, with the orchestration system providing additional gains on top of that baseline performance.

Cost analysis was a key part of the evaluation. When running GPT-5 alone with a basic prompt, the system achieved 78.9% accuracy at an estimated cost of $8,400 per case. Adding the DxO orchestration reduced the cost by approximately $3,000 while improving accuracy by about four points. However, on emergency department cases, the system showed little improvement in either accuracy or cost efficiency compared to the standalone model. With the newer GPT-5.5 model, released in April 2026, the multi-agent approach boosted accuracy on complex cases from 72% to nearly 87% at a slightly lower cost, though emergency department performance remained similar.

The benefits of the orchestration system are most evident when paired with less powerful models. For example, Google’s Gemini 2.5 Flash model scored 46.4% on its own but rose to nearly 69% when assisted by DxO. Despite this boost, the gap between weaker and stronger models remained larger than the improvement provided by the orchestration layer. Researchers noted that accuracy gains naturally diminish as models approach theoretical limits, but the system helps optimize both cost and performance.

Suleyman stated that Microsoft is currently collaborating with the Mayo Clinic to train a foundation model specifically for healthcare, with a release planned for next year. He believes this new model will allow for massive leaps in capability. He predicts that by 2027, the company will be on the verge of a breakthrough in medical record coding, enabling clinicians to stream highly accurate predictions about how a patient’s condition might unfold. Suleyman, who previously co-founded DeepMind and joined Microsoft in 2024, sees this work as a continuation of his career-long focus on using AI to solve medical challenges.

The next phase involves real-world validation. Microsoft has partnered with Beth Israel Deaconess Medical Center in Boston to evaluate the clinical reasoning capabilities of DxO across various hospital settings. This study will compare the AI’s recommendations against those of human doctors. Patients will receive standard care, including triage and specialist consultations, while those who consent will also engage with the AI system. Dominic King, vice president of Health at Microsoft AI, described the initiative as the largest study of its kind, aiming to provide compelling evidence for the future deployment of these systems at scale.

This move places Microsoft in direct competition with Google and OpenAI, both of which have recently claimed that their AI tools can match or outperform human doctors. Google published a similar study in The Lancet this month evaluating its AMIE diagnostic AI in urgent care settings. Microsoft’s new study aligns with a broader manifesto outlining the requirements for medical superintelligence, which is defined as the capacity to surpass human-led health systems in reasoning and coordination. The company aims to move beyond simple benchmarks and integrate deeply with clinical workflows to improve patient outcomes.

Microsoft already sells dictation and radiology tools to clinicians and views DxO as an expansion of its existing portfolio. Suleyman emphasized the company’s commitment to building the number one state-of-the-art medical clinical decision support product in the coming years. He expressed confidence that progress will accelerate quickly once a production flywheel is established, similar to the rapid advancements seen in coding and chatbot development. The goal is to identify gaps in real-world usage and iterate on the model to improve its performance continuously.