A recent study led by researchers from Harvard Medical School, published in the journal Science, indicates that OpenAI's advanced reasoning model significantly surpasses experienced human doctors in diagnosing rare diseases within emergency settings. The research tested the model on 76 real-life emergency room cases from a Boston hospital, where it was tasked with diagnosing conditions and creating testing plans. The AI's performance was evaluated at three critical stages: initial triage, first doctor consultation, and final admission decisions, consistently showing results equal to or better than those of expert physicians. Notably, the AI excelled during the initial triage phase, where it had the least information available, and proved effective in identifying rare and complex medical conditions. The model's clinical decision-making capabilities, such as suggesting treatments, outperformed previous AI systems and human doctors. However, the study's authors, including senior co-author Arjun Manrai, emphasized that AI should not replace doctors, as real-world physicians utilize a broader range of data, including patient interactions and imaging studies. The study's implications suggest that AI could assist in triage and provide reliable second opinions, with a 2025 study indicating that one in five clinicians are already seeking AI for such support.