Microsoft AI Outperforms General Physicians in Complex Medical Diagnosis Simulations

For patients in Texas and across the South Central region, the future of emergency room triage may look less like a wait for a specialist and more like a rapid, AI-assisted assessment. A new diagnostic system developed by Microsoft has demonstrated a significant advantage over human generalists in handling complex medical scenarios, according to a study published in Science on Thursday. The system, known as MAI-DxO, correctly identified diagnoses in difficult cases more than twice as often as primary care doctors and general internists working individually.

The architecture of MAI-DxO mimics a panel of physicians rather than a single clinician. It coordinates multiple artificial intelligence agents that simulate a multidisciplinary medical team. These agents engage in a structured process that includes questioning the patient, ordering specific diagnostic tests, and debating potential diagnoses before reaching a final conclusion. This approach is designed to mirror the collaborative nature of high-stakes medical decision-making.

In the simulation, which utilized complex cases from the New England Journal of Medicine, the AI system achieved an 83% accuracy rate. By comparison, human generalists using reference materials were correct only 36% of the time. It is important to note that these highly complex cases are typically managed by specialists or teams in real-world settings, rather than the generalists included in this specific comparison. Dominic King, Vice President of Health at Microsoft AI, described the findings as substantial, noting that the current healthcare system faces a growing mismatch between the demand for expertise and the available supply.

The study highlights that much of the performance gain stems from the underlying power of the AI models used, with Microsoft’s orchestration system providing additional efficiency. Microsoft tested the system with models from several major tech firms, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The highest accuracy scores were achieved using OpenAI’s GPT-5 and GPT-5.5 models. However, the economic and accuracy benefits varied significantly depending on the model and the type of case.

Cost and accuracy metrics showed mixed results. When using GPT-5 alone, the system cost approximately $8,400 per complex case with a 78.9% accuracy rate. Adding the DxO orchestration reduced the cost by about $3,000 but only increased accuracy by four points. For emergency department cases, the improvement was minimal. With the newer GPT-5.5 model, however, DxO boosted accuracy from 72% to nearly 87% for complex cases while slightly reducing costs. The gains were less pronounced for weaker models, suggesting that the orchestration layer is most beneficial when paired with high-performance underlying engines.

Mustafa Suleyman, CEO of Microsoft AI, framed these results as a step toward “medical superintelligence,” a concept defined as surpassing human-led health systems in medical reasoning and coordination. Suleyman emphasized that Microsoft is collaborating with the Mayo Clinic to develop a foundation model for healthcare, with a release planned for next year. He predicted that by 2027, the company will achieve a breakthrough in “coding style” for medical records, allowing for real-time streaming of accurate predictions about patient outcomes.

Beyond the laboratory, Microsoft is preparing for real-world validation. Through a new partnership with Beth Israel Deaconess Medical Center in Boston, the company will evaluate DxO’s clinical reasoning in various hospital settings, including emergency and urgent care environments. Patients will receive standard care while consenting to parallel engagement with the AI system, allowing researchers to compare recommendations. This initiative follows a similar study by Google and aims to provide compelling evidence for the large-scale deployment of AI in clinical decision support.

Suleyman, who previously co-founded DeepMind and led its health initiatives, believes that healthcare models will see an order of magnitude improvement in capability within the next 24 to 36 months. He views this period as a critical flywheel for medical records and clinical support, similar to the rapid progress seen in coding and chat applications. Microsoft aims to position itself as the leader in state-of-the-art medical clinical decision support, building on its existing sales of dictation and radiology tools to clinicians.