AI Diagnostics: Microsoft claims its system outperforms generalist physicians on complex cases

For patients navigating complex health issues, the landscape of diagnostic support is shifting. A study published this week in Science indicates that Microsoft’s new AI diagnostic system, MAI-DxO, correctly identified the nature of difficult medical cases in simulations more than twice as often as generalist physicians. While this does not yet replace the human clinician, it signals a potential change in how medical expertise is scaled to meet growing demand, offering a new layer of analytical support in an environment where supply of specialists often lags behind patient needs.

The performance gap is stark in the data. In a simulation utilizing complex cases from the New England Journal of Medicine, the AI system was accurate 83% of the time. In contrast, primary care doctors and general internists, working individually with standard reference materials, were correct only 36% of the time. It is important to note that such complex cases are typically managed by specialists or teams in real-world settings, who were not included in this specific comparison. This suggests the AI is currently best positioned to assist or challenge initial assessments made by generalists.

The system, known as DxO, operates by coordinating multiple AI agents that function similarly to a panel of physicians. These agents interact by questioning the patient, ordering necessary tests, and debating potential diagnoses. Dominic King, vice president of Health at Microsoft AI, described the findings as "very substantial and exciting," citing the current mismatch between the demand for medical expertise and its supply as the primary driver for this technology. By joining Google and OpenAI in claiming AI parity or superiority in specific diagnostic tasks, Microsoft is entering a crowded field of competitors vying for dominance in health tech.

However, the study reveals that much of the accuracy gain stems from the underlying AI models rather than Microsoft’s orchestration system alone. Microsoft tested DxO with models from OpenAI, Google, Anthropic, Meta, and DeepSeek, finding the best results with OpenAI’s GPT-5 and GPT-5.5. When using GPT-5 with a basic prompt, the accuracy was 78.9% at a cost of approximately $8,400 per case. Adding the DxO orchestration reduced the cost by about $3,000 but only increased accuracy by four points. For emergency department cases, the improvement in both cost and accuracy was negligible when comparing the standalone model to the orchestrated system.

The dynamics shift slightly with newer models. Using GPT-5.5, released in April 2026, the DxO system boosted accuracy on complex cases from 72% to nearly 87% at a slightly lower cost. Yet, this advantage did not extend to emergency department cases, where both versions scored around 85% at similar costs. The system proved most beneficial for less capable models; for instance, Gemini 2.5 Flash saw its accuracy rise from 46.4% to nearly 69% with DxO. Nevertheless, the inherent gap between weaker and stronger models remained larger than the boost provided by the orchestration layer.

Mustafa Suleyman, CEO of Microsoft AI, frames these developments as a step toward "medical superintelligence." He noted that Microsoft is collaborating with the Mayo Clinic to train a foundation model for healthcare, with a release planned for next year. Suleyman predicts that by 2027, the industry will see a "coding style breakthrough" in medical records, allowing for highly accurate predictions of how a patient's condition might unfold. He views DxO as part of the foundational infrastructure required to achieve these leaps, aiming to provide clinicians with foresight he describes as "gold dust."

Suleyman, a co-founder of DeepMind and former head of DeepMind Health, believes generative AI in healthcare is still in its early stages. He anticipates that healthcare models will improve by an order of magnitude in capability within the next 24 to 36 months. He points to the "flywheel" effect seen in coding and chatbots, where models are iterated upon using real-world usage to identify and patch gaps. He expects this cycle to begin for medical records and clinical decision support next year, driving rapid progress.

To validate these capabilities in real-world settings, Microsoft has partnered with Beth Israel Deaconess Medical Center in Boston. This evaluation will assess DxO’s clinical reasoning across various hospital environments. Patients will receive standard treatment, including triage and specialist consultation, while consenting participants will also interact with the AI system. Microsoft and its partners will then compare the AI’s recommendations against those of human doctors. King called this initiative "the biggest study of its kind," noting it will cover both emergency and urgent care settings.

This study coincides with a broader manifesto from Microsoft defining "medical superintelligence" as the capacity to surpass human-led health systems in reasoning and coordination. The company aims to move beyond benchmarking individual tasks to integrating with clinicians for better overall outcomes. As Microsoft already sells dictation and radiology tools, DxO represents an expansion of its clinical portfolio. Suleyman stated that the company is committed to having the "absolute number one state-of-the-art medical clinical decision support products" in the coming years, positioning this focus as a key differentiator among frontier AI labs.