Microsoft AI Claims Superiority Over Physicians in Complex Diagnostics Study

Microsoft’s AI unit has released a diagnostic system that appears to outperform generalist physicians in handling complex medical cases, a result the company’s leadership is framing as a critical step toward "medical superintelligence." A study published this week in the journal Science details how this system, known as MAI-DxO, correctly identified diagnoses in simulated scenarios more than twice as often as primary care doctors and general internists. Mustafa Suleyman, the CEO of Microsoft AI, presented these findings as evidence that the technology is moving beyond simple assistance to becoming a core component of medical reasoning.

The system operates by coordinating multiple AI agents that function similarly to a multidisciplinary medical panel. Instead of relying on a single algorithm, these agents interact by questioning virtual patients, ordering diagnostic tests, and debating potential diagnoses before reaching a conclusion. In the specific simulation, which utilized complex cases previously published in the New England Journal of Medicine, the AI system achieved an 83% accuracy rate. By comparison, human physicians working individually with access to reference materials were correct in only 36% of the instances. However, a key caveat noted in the report is that these high-complexity cases are typically managed by specialists or teams in real-world settings, groups that were not included in the direct comparison.

The publication places Microsoft in direct competition with Google and OpenAI, both of which have recently claimed that their AI capabilities can match or exceed human medical expertise. Dominic King, vice president of Health at Microsoft AI, described the findings as "very substantial and exciting." He highlighted that the primary challenge in modern healthcare is the widening gap between the demand for medical expertise and the available supply. King suggested that this technology addresses that specific mismatch by providing a scalable source of diagnostic reasoning.

A deeper look at the study’s data suggests that the performance gains are driven primarily by the power of the underlying large language models, with the multi-agent orchestration providing an additional layer of improvement. Microsoft tested the system using models from several major providers, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The most favorable results were achieved when the system utilized OpenAI’s GPT-5 and GPT-5.5 models. When GPT-5 was used independently with a basic prompt, it scored 78.9% on complex cases at an estimated cost of $8,400 per case. The addition of the DxO orchestration reduced the cost by roughly $3,000 but improved accuracy by only about four percentage points.

The benefits of the orchestration layer were less consistent across different case types. For emergency department scenarios, there was minimal difference in accuracy or cost between using the base model alone and using it with the DxO system. However, with the newer GPT-5.5 model, released in April 2026, the results diverged. On complex cases, the DxO system boosted GPT-5.5’s accuracy from 72% to nearly 87% at a slightly lower cost. Yet, in emergency cases, both configurations scored around 85%, indicating that the added complexity of the multi-agent system offers diminishing returns in high-volume, acute settings.

The study also revealed that the orchestration system is more effective when paired with less powerful models. For instance, Google’s Gemini 2.5 Flash model scored 46.4% independently 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. Microsoft researchers noted that as models approach the theoretical maximum of performance, the marginal gains from adding coordination layers naturally shrink. A company spokesperson stated that the system helps optimize the trade-off between accuracy and operational costs.

Looking ahead, Suleyman indicated that Microsoft is collaborating with the Mayo Clinic to develop a specialized foundation model for healthcare, with a release expected next year. He believes this dedicated model will allow for "really enormous leaps" in capability. Suleyman is particularly focused on the potential for a "coding style breakthrough" in medical records by 2027, envisioning a system that can stream highly accurate predictions about how a patient’s condition might evolve. He described this foresight as "gold dust" for clinicians, allowing them to anticipate outcomes rather than just react to them.

To validate these capabilities in real-world environments, Microsoft has partnered with Beth Israel Deaconess Medical Center in Boston. This collaboration will evaluate DxO’s clinical reasoning across various hospital settings, including emergency and urgent care. The study design involves patients receiving standard care while consenting participants also interact with the AI system. Researchers will then compare the AI’s recommendations against the decisions made by human doctors. King called this initiative the "biggest study of its kind," noting that it will provide compelling evidence on the scalability of such systems in diverse clinical contexts.

This move coincides with a broader manifesto from Microsoft defining "medical superintelligence" as the ability to surpass the strongest human-led health systems in reasoning and coordination across the entire arc of care. The company aims to move beyond benchmarking individual tasks, such as answering exam questions, toward deep integration with clinicians and healthcare infrastructure. Suleyman emphasized that this focus on real-world clinical decision support distinguishes Microsoft from other frontier labs. He stated that the goal is to establish the "absolute number one state-of-the-art" medical product in the coming years, leveraging the company’s existing portfolio of dictation and radiology tools.