Microsoft Says Its AI Diagnosed Complex Cases More Than Twice as Accurately as Doctors as Suleyman Chases "Medical Superintelligence"

Microsoft says its artificial intelligence system can diagnose complex medical cases far more accurately than generalist doctors, a result its AI chief Mustafa Suleyman described as a step towards "medical superintelligence."

In a simulation study published on Thursday, October 22, in the journal Science, Microsoft's AI diagnostic system, MAI-DxO, correctly diagnosed 83% of complex cases drawn from the New England Journal of Medicine, TechCrunch reported. Primary care doctors and general internists working alone with reference materials got 36% right.

The comparison has an important caveat. Complex cases of this kind are usually handled by specialists or teams of physicians, who were not part of the study. Nonetheless, the results add to growing evidence that AI systems can perform sophisticated clinical reasoning.

How MAI-DxO works

MAI-DxO, or DxO, is designed to mimic a panel of doctors. It coordinates several AI agents that question the patient, order tests and debate possible diagnoses before reaching a conclusion. This approach, known as orchestration, aims to replicate the way experienced clinicians work through difficult cases step by step.

Microsoft tested DxO with AI models from several companies, including OpenAI, Google, Anthropic, Meta and DeepSeek. Its best results came with OpenAI's GPT-5 and GPT-5.5.

The findings show that the underlying model matters a great deal. On complex cases, GPT-5 on its own, with a basic prompt, achieved 78.9% accuracy at a cost of about $8,400 per case in simulated tests and procedures. Adding DxO cut costs by about $3,000 while raising accuracy by about four percentage points. With GPT-5.5, released in April 2026, DxO raised accuracy on complex cases from 72% to nearly 87% at slightly lower cost.

For a weaker model, Google's Gemini 2.5 Flash, accuracy rose from 46.4% alone to nearly 69% with DxO. But the gap between weaker and stronger models was still larger than DxO's boost. On emergency department cases, the gains were smaller, with both setups scoring about 85% using GPT-5.5.

"True believers"

Suleyman, who co-founded DeepMind, the AI lab Google acquired in 2014, and joined Microsoft in 2024 to lead its consumer AI efforts, made clear his ambitions for healthcare. "We're just true believers," he said. He predicted that healthcare AI models would be "an order of magnitude better in capability within the next 24 to 36 months."

Dominic King, vice-president of health at Microsoft AI, who previously worked with Suleyman at DeepMind Health, said: "There is something very substantial and exciting here." He argued that AI could help address a fundamental problem: "The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply."

From simulation to the clinic

Simulation studies are only a first step. Microsoft announced a partnership with Beth Israel Deaconess Medical Center in Boston to evaluate DxO's clinical reasoning against human doctors in real emergency and urgent care settings. It is also working with Mayo Clinic on a healthcare foundation model, planned for release next year.

Google, Microsoft's main rival in medical AI, published a related study of its AMIE diagnostic AI with Beth Israel in The Lancet, underlining the competition between technology giants to lead in clinical AI.

Microsoft already sells AI tools to clinicians for dictation and radiology, and its Nuance unit provides speech recognition widely used by doctors.

The caveats

Medical experts caution that strong performance in simulations does not guarantee safe and effective use in practice. Real patients present with incomplete information, emotional needs and complex social circumstances. Diagnoses must be communicated, treatment decisions shared, and responsibility clearly assigned.

Regulators will also play a key role. AI systems used for diagnosis are likely to be treated as medical devices in many countries, requiring evidence of safety and effectiveness. Questions about liability, data privacy and bias in training data remain unresolved.

What it means for India

India faces one of the world's starkest mismatches between healthcare demand and supply. The country has a shortage of doctors, particularly specialists, and they are heavily concentrated in cities. Patients in rural areas and smaller towns often lack access to specialists for complex conditions.

AI diagnostic tools could help bridge that gap by supporting general practitioners and health workers in making better decisions, and by prioritising which patients need specialist referrals. Indian startups and hospitals are already experimenting with AI for screening and diagnostics, and companies such as CureBay, which raised ₹100 crore this week, are using AI to support rural clinics.

However, deploying such tools in India requires validation on Indian patient data, integration with local health systems and regulatory clarity. The cost of AI systems, which Microsoft's study measured in simulated test expenses, must also be weighed against the resources available in low-income settings.

The race in medical AI

Technology companies see healthcare as one of the largest opportunities for AI. Google has developed medical AI models and systems such as AMIE for diagnostic conversations, while OpenAI has promoted the use of its models for health questions. Startups focused on clinical documentation, diagnostics and drug discovery have attracted significant funding.

Hospitals and health systems have begun adopting AI for administrative tasks, such as generating clinical notes from conversations between doctors and patients, which can reduce paperwork and burnout. Diagnostic applications, which directly influence patient care, face higher bars for evidence and regulation.

The Science publication gives Microsoft's work scientific credibility through peer review, an important step as companies seek to persuade clinicians and regulators that their systems are reliable.

What comes next

The Beth Israel evaluation will provide the first real-world evidence of how DxO performs alongside human doctors. The results will be closely watched by hospitals, regulators and investors.

For Microsoft, healthcare is a strategic priority, offering a large market and a compelling demonstration of AI's potential. For patients, the promise is faster and more accurate diagnosis. The challenge will be ensuring that promise is delivered safely. The question for policymakers will be how to harness such tools to expand access to quality care, particularly in countries like India, while ensuring patient safety, accountability and equity. Getting that balance right could determine whether AI narrows or widens existing gaps in healthcare. The coming year of clinical evaluations will be pivotal in testing those claims against reality.