Microsoft’s Mustafa Suleyman says its AI outperforms doctors on hard cases as he chases ‘medical superintelligence’

Microsoft is leaning hard into healthcare, and it has a new benchmark result to wave around. In a study published Thursday in Science, its diagnostic system correctly handled complex medical cases in simulation 83% of the time, more than double the 36% score posted by primary care doctors and general internists working with reference materials.

The system is called MAI-DxO, short for DxO, and it works less like a single model than a little committee. Several AI agents question the patient, request tests, and argue over the diagnosis. The setup was tested on difficult cases from the New England Journal of Medicine. One caveat sits right there in the fine print: these are the kinds of cases usually handled by specialists or physician teams, and they were not included in the comparison.

Microsoft says the gains mostly come from stronger base models, with its orchestration layer adding another push on top. It tested MAI-DxO with models from OpenAI, Google, Anthropic, Meta, and DeepSeek, but the best results came with OpenAI’s GPT-5 and GPT-5.5. On its own, GPT-5 reached 78.9% on complex cases at roughly $8,400 per case. With DxO, Microsoft says the cost fell by about $3,000 while accuracy rose only about four points. The story was different with GPT-5.5: on complex cases, DxO lifted accuracy from 72% to nearly 87% at a slightly lower cost.

The pattern breaks down outside the hardest cases. On emergency department cases, DxO barely moved the needle for either accuracy or cost. Microsoft says the system helps most with complex cases and weaker models, such as Gemini 2.5 Flash, which went from 46.4% alone to nearly 69% with DxO. But even there, the stronger model still mattered more than the orchestration around it.

Mustafa Suleyman, who leads Microsoft AI, is calling this a step toward “medical superintelligence,” and he’s not exactly being subtle about the company’s ambitions. Microsoft says it is working with the Mayo Clinic on a healthcare foundation model slated for release next year, and it has a new partnership with Beth Israel Deaconess Medical Center in Boston to test DxO in real hospital settings. Google is running a similar clinical study with Beth Israel this month. The race now is not just to beat doctors on slides, but to prove any of this helps in the messy business of actual care.