A new AI system called DeepRare has outperformed experienced physicians in diagnosing rare diseases, according to a study published in Nature. The breakthrough marks a significant advancement in medical artificial intelligence, particularly in tackling one of healthcare’s most challenging problems: the delayed and often inaccurate diagnosis of rare conditions. Rare diseases affect an estimated 300 million people worldwide, yet they remain difficult to identify due to their complex and varied symptoms, which often mimic more common illnesses. On average, patients endure a diagnostic journey lasting five years or longer, involving multiple specialist visits, misdiagnoses, and unnecessary treatments. To address this, researchers developed DeepRare, an innovative agentic AI framework that uses a team of 40 specialized digital tools working in concert under a central AI coordinator. Unlike traditional AI models that attempt to solve problems in isolation, DeepRare leverages a collaborative approach—analyzing diverse data sources including genetic sequences, medical literature, electronic health records, and even handwritten clinical notes. In initial testing, DeepRare was evaluated on 6,401 previously diagnosed cases. When presented with the same symptoms and genomic data that doctors had access to years earlier, the AI was able to identify the correct rare disease significantly faster and with greater accuracy than the original clinical teams. It also surpassed 15 other existing diagnostic tools. The true test came in a head-to-head comparison involving 163 particularly complex cases. Five expert physicians, each with over ten years of experience in rare disease diagnosis, were given the same data as DeepRare. The AI correctly identified the disease on its first attempt 64.4% of the time, compared to 54.6% for the doctors. “DeepRare is one of the first computational models to surpass the diagnostic performance of expert physicians in the complex task of rare-disease phenotyping and diagnosis,” the research team stated in their paper. Even when DeepRare did not get the correct answer immediately, it remained highly effective—achieving a high Recall@3 score, meaning the right diagnosis was almost always among its top three suggestions. Ten rare disease specialists reviewed the AI’s step-by-step reasoning process and agreed with its logic in 95.4% of cases. The study highlights the transformative potential of large-language-model-driven agentic systems in clinical settings. The researchers believe DeepRare could not only improve diagnostic accuracy but also streamline clinical workflows, reduce patient suffering, and shorten the time to diagnosis. The findings were accompanied by a commentary article in Nature, underscoring the broader implications of this technology for medicine and AI integration in healthcare. The work represents a milestone in using AI not just as a tool, but as a collaborative partner in complex medical decision-making.