Microsoft's AI diagnostic system correctly resolved complex medical cases more than twice as often as individual generalist physicians in a simulation, according to a study published Thursday in Science. Mustafa Suleyman, CEO of Microsoft AI, framed the result as a step toward what the company calls "medical superintelligence." The headline number — 83% accuracy for the system, dubbed MAI-DxO, against 36% for primary care doctors and general internists working alone with reference materials — is striking, but the study's own design and data complicate any simple reading that AI has surpassed doctors.

How the system works

MAI-DxO, short for Microsoft AI Diagnostic Orchestrator, does not rely on a single model. It coordinates several AI agents that behave like a panel of physicians: they question a simulated patient, order tests, and debate a diagnosis before converging on an answer. The cases came from the New England Journal of Medicine's complex clinical challenges, a benchmark built around rare and tangled presentations rather than routine complaints. Microsoft tested the orchestrator with models from OpenAI, Google, Anthropic, Meta, and DeepSeek, and its best results came with OpenAI's GPT-5 and GPT-5.5.

What the data actually shows

The study measured cost alongside accuracy, estimating the expense of running the AI on each case. GPT-5 alone, with a short basic prompt, scored 78.9% on complex cases at roughly $8,400 per case. Adding DxO cut that cost by about $3,000 but improved accuracy by only around four points. On emergency department cases, neither accuracy nor cost improved meaningfully with the orchestration layer. With GPT-5.5, released in April 2026, the results split: on complex cases DxO lifted accuracy from 72% to nearly 87% at slightly lower cost, while emergency cases stayed flat at about 85% either way. The pattern suggests much of the gain comes from stronger underlying models, with orchestration adding incremental value on top. DxO helped weaker models more — Gemini 2.5 Flash rose from 46.4% alone to nearly 69% with DxO — but the gap between weak and strong models remained larger than the boost DxO provided. Microsoft's researchers attributed the shrinking gains to models approaching a theoretical ceiling.

The comparison gap

The 36% figure belongs to primary care doctors and general internists working individually. Cases of this complexity are typically handled by specialists or multidisciplinary physician teams, and those clinicians were not part of the comparison. That omission matters for anyone reading the 83% versus 36% contrast as a verdict on human medicine. Dominic King, vice president of Health at Microsoft AI, told TechCrunch that the core problem is "the growing mismatch between demand and supply of medical expertise" — a framing about access and throughput rather than replacement.

Microsoft's roadmap

Suleyman said Microsoft is working with Mayo Clinic to train a foundation model for healthcare, with release planned for next year, and he expects the company to "make really enormous leaps here." He predicted that 2027 will bring a coding-style breakthrough for medical records, with systems streaming accurate predictions about how a record might unfold. He also said healthcare models should get "an order of magnitude better in capability within the next 24 to 36 months," driven by a flywheel of production usage feeding back into model improvement. The company already sells dictation and radiology tools to clinicians and positions DxO as an expansion of that line.

The real-world test ahead

The more consequential evidence will come from a new partnership with Beth Israel Deaconess Medical Center in Boston, where Microsoft will evaluate DxO's clinical reasoning across hospital settings. Patients will receive standard care — triage, specialist time, treatments, investigations, discharge decisions — while consenting patients also engage with DxO in parallel, and researchers compare its recommendations against the human doctors'. King called it likely "the biggest study of its kind," spanning emergency and urgent care environments. Google, meanwhile, published a clinical study with Beth Israel in The Lancet this month evaluating its AMIE conversational diagnostic AI in real-world urgent care visits with physicians monitoring live. Microsoft's Science paper joins a growing set of vendor-authored claims that their AI can match or outperform doctors; independent, prospective clinical evidence remains the measure that will decide whether the 83% figure translates into better care.