Microsoft AI claims to beat doctors on complex cases

For patients facing difficult medical questions, a new AI system from Microsoft is claiming to offer a significant advantage over traditional generalist care. In a study published this week in the journal Science, the system, known as MAI-DxO, correctly identified complex medical conditions in a simulated environment more than twice as often as human doctors working individually. While the technology is not yet in widespread clinical use, the results suggest a shift in how artificial intelligence might assist in challenging diagnostic scenarios.

The system functions as a coordinated panel of AI agents rather than a single chatbot. These agents work together to question virtual patients, order necessary tests, and debate potential diagnoses, mimicking the collaborative process of a medical team. In the simulation, which used complex cases from the New England Journal of Medicine, the AI panel achieved an accuracy rate of 83 percent. By comparison, primary care doctors and general internists, who had access to reference materials but worked alone, were correct only 36 percent of the time. It is important to note that such complex cases are typically handled by specialists or teams in real-world settings, groups that were not included in this specific comparison.

Microsoft AI CEO Mustafa Suleyman described the results as a step toward what he calls 'medical superintelligence.' He stated that the company is working with the Mayo Clinic to train a specific foundation model for healthcare, with a release planned for next year. Suleyman believes this infrastructure will allow for 'really enormous leaps' in the field. He also predicted that by 2027, there will be a breakthrough in 'coding style' for medical records, enabling highly accurate predictions about how a patient's condition might unfold over time.

However, the study’s data indicates that much of the performance improvement comes from the power of the underlying AI models rather than the coordination system alone. Microsoft tested the system using models from several companies, including OpenAI, Google, Anthropic, Meta, and DeepSeek. The best results were achieved when the system was paired with OpenAI’s GPT-5 and GPT-5.5 models. When using GPT-5 on its own with a basic prompt, the system scored 78.9 percent on complex cases at an estimated cost of about $8,400 per case. The MAI-DxO orchestration reduced that cost by roughly $3,000 but only increased accuracy by about four points.

With the newer GPT-5.5 model, released in April 2026, the benefits of the coordination system were more pronounced for complex cases. The system raised the accuracy score from 72 percent to nearly 87 percent at a slightly lower cost. Yet, for emergency department cases, the gains were minimal, with or without the additional layer of coordination, both scoring around 85 percent. The system proved most helpful when paired with less capable models; for example, Gemini 2.5 Flash saw its accuracy jump from 46.4 percent to nearly 69 percent when assisted by the orchestration system, though the gap between weak and strong models remained larger than the boost provided by the tool.

Microsoft is moving beyond simulations to real-world testing through a new partnership with Beth Israel Deaconess Medical Center in Boston. This evaluation will assess the system's clinical reasoning across various hospital settings, including emergency and urgent care environments. Patients will receive standard treatment while those who consent will also engage with the AI. The goal is to compare the AI's recommendations directly against those of human doctors. Dominic King, Microsoft AI’s vice president of Health, noted that this will be one of the largest studies of its kind, aiming to provide compelling evidence for the potential deployment of these systems at scale.

This research coincides with a broader manifesto from Microsoft defining 'medical superintelligence' as the ability to surpass current human-led health systems in reasoning and coordination. The company aims to move beyond simple benchmarking tasks like answering exam questions and instead integrate AI deeply with clinicians and healthcare systems. Suleyman, who previously co-founded DeepMind and joined Microsoft in 2024, emphasized that progress in this field often follows a 'flywheel' effect, where models improve rapidly through real-world usage and iteration. He expects healthcare models to become an order of magnitude more capable within the next 24 to 36 months.

Microsoft already sells dictation and radiology tools to clinicians and views this new diagnostic capability as an expansion of those efforts. Suleyman stated that the company is committed to accelerating the introduction of this technology to the world. He believes that having the state-of-the-art medical decision support products will be a key differentiator for Microsoft compared to other frontier AI labs. The company is positioning itself to lead in this space as the technology moves from laboratory simulations to daily clinical practice.