83%: The accuracy rate Microsoft AI achieved in complex medical simulations

An accuracy rate of 83% stands out in the latest data from Microsoft AI, which demonstrated that its diagnostic system, MAI-DxO, correctly identified complex medical cases in a simulation more than twice as often as generalist physicians. The results, published Thursday in the journal Science, show the system surpassing individual doctors who worked with reference materials and achieved a 36% success rate. Mustafa Suleyman, CEO of Microsoft AI, described this milestone as a critical step toward achieving what he terms medical superintelligence, a capacity that surpasses current human-led health systems in reasoning and coordination.

The system functions as a coordinated panel of AI agents that mimic a group of doctors. These agents question the patient, order necessary tests, and debate the final diagnosis. The study utilized complex cases from the New England Journal of Medicine, though researchers noted that such difficult scenarios are typically handled by specialists or teams rather than the individual generalists used for comparison. Dominic King, Vice President of Health at Microsoft AI, emphasized the significance of the gap, noting a growing mismatch between the demand for medical expertise and its supply.

Reactions to the findings highlight the competitive landscape, as Microsoft joins Google and OpenAI in claiming that AI can match or outperform medical professionals. However, the data suggests that much of the performance improvement stems from powerful underlying models rather than just the orchestration system. Microsoft tested MAI-DxO with models from various companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The highest results were recorded using OpenAI's GPT-5 and GPT-5.5 models. With GPT-5.5, released in April 2026, the system boosted accuracy on complex cases from 72% to nearly 87% while slightly reducing costs.

Cost and efficiency were also key metrics in the evaluation. When using GPT-5 alone with basic prompts, the system scored 78.9% on complex cases at an estimated cost of $8,400 per case. Adding the DxO orchestration reduced that cost by roughly $3,000 while increasing accuracy by only about four points. On emergency department cases, the gains in both accuracy and cost were minimal. The system proved most helpful when paired with less capable models, such as Gemini 2.5 Flash, which saw its score rise from 46.4% to nearly 69% with the assistance of DxO.

Looking ahead, Microsoft is partnering with the Mayo Clinic to train a healthcare foundation model, with a release planned for next year. Suleyman believes this model will enable significant leaps in the field. He also shared a vision for 2027, predicting a coding style breakthrough for medical records that will allow for streaming accurate predictions about how a record might unfold. This foresight, he argued, is invaluable for clinicians.

To validate these capabilities in real-world settings, Microsoft has established a new partnership with Beth Israel Deaconess Medical Center in Boston. The evaluation will assess clinical reasoning across different hospital environments, including urgent care and emergency departments. Patients will receive standard treatment while consenting individuals will also interact with DxO. The team will then compare the AI's recommendations against those of human doctors. King expressed confidence that this will be the largest study of its kind, providing compelling evidence for deploying these systems at a greater scale in the future.

Suleyman, a co-founder of DeepMind and former head of Google DeepMind Health, has long championed the use of AI in medicine. He expects healthcare models to become an order of magnitude better in capability within the next 24 to 36 months. Microsoft already sells dictation and radiology tools to clinicians and views DxO as an expansion of these efforts. The company aims to have the top state-of-the-art medical clinical decision support products over the next few years, positioning this focus as a key differentiator among frontier AI labs.