Microsoft claims its AI tool outperforms doctors in disease diagnosis.

For years, leaders in AI have pointed to healthcare as a prime example of how their technology can enhance lives. Now, according to Microsoft AI CEO Mustafa Suleyman, “it is finally here.” The company claims to have created an AI-driven tool that can diagnose medical conditions with up to four times the accuracy of human doctors.

Recent findings, published in a top-tier scientific journal, showcase tests where Microsoft’s AI system was compared to human physicians across various scenarios derived from medical case reports and emergency department data. Microsoft has compiled these cases into a benchmark called SDBench, which it intends to make open-source. The Diagnostic Orchestrator, designed to emulate a panel of medical experts, utilizes multiple AI agents to analyze symptoms and medical histories, determine necessary tests, and arrive at a diagnosis, mirroring the investigative process of doctors.

In tests using cases from numerous reports in a leading medical journal, Microsoft’s system achieved about 83 percent diagnostic accuracy, while generalist physicians scored around 36 percent with references and only 22 percent without. The AI also excelled in emergency department scenarios from a large patient admission database, achieving 80 percent accuracy compared to an average of 47 percent for human doctors. Additionally, the AI system often provided diagnoses at a lower estimated cost than its human counterparts.

The researchers noted that the system, powered by OpenAI’s GPT-5, outperformed other advanced AI models operating independently, although the peer-reviewed study did not include the latest models like GPT-6 Astra or Google’s Gemini 4 Argon. Initially developed when GPT-4.1 was the leading model, subsequent tests indicate that “improvements in model capability do not necessarily diminish the benefits” of collaborative AI agents.

Microsoft recognizes that there’s still a long way to go before this technology can be widely implemented. While the tests simulate the diagnostic process of doctors—asking questions, considering possible explanations, and ordering tests—the system has only been evaluated on historical medical cases, leaving its performance in real clinical settings uncertain. “It needs clinical validation,” Suleyman states. “That is a leap, but this is a major milestone along that journey.”

This research is part of Microsoft’s broader goal to create what it terms “medical superintelligence”—AI that can surpass the best human-led healthcare systems. The term “superintelligence” has been part of Microsoft’s vocabulary long before it became a buzzword in political circles.

Diagnosis is just one aspect of Microsoft’s healthcare vision, according to Dr. Dominic King, the company’s AI health lead and coauthor of the study. In a recent blog post and an article in a medical journal, Microsoft outlined five additional areas where it believes AI must excel, including coordinating clinical care, making treatment decisions, and addressing the human aspects of medicine.

King mentioned that Microsoft plans to start testing the system in hospital environments with the Beth Israel Deaconess Medical Center to evaluate its performance in real-world clinical settings. While neither King nor Suleyman provided a specific timeline, King indicated that testing would likely commence in the next few months.

Commercialization may also be on the horizon, although Microsoft has not disclosed when or how the technology might be made available. The company already has a strong foothold in U.S. healthcare through its digital tools, and health applications are among the most popular uses for its consumer-facing chatbot. King stressed the importance of a responsible rollout, given the stringent regulatory and compliance requirements in the medical field.

“Next year we’re going to see extremely impressive deployments of intelligent systems in production,” Suleyman predicts. “I think this is just going to change the way care is delivered in a big way.”

Suleyman has previously shared his “human-centered” vision for AI, emphasizing that medicine is no exception. While he anticipates the emergence of “something that is indistinguishable from human performance on medical superintelligence in the next couple of years,” he insists that human clinicians will always play a vital role in patient care.

He acknowledges that AI might lead to doctors taking on more tasks, but he hopes it will alleviate some of “the drudgery” in medicine, allowing clinicians to spend more quality time with patients.

“I’m very optimistic about what’s coming next,” he concludes. “I think this is why we build technology.”