An analysis of large language model (LLM) diagnostic performance on medical benchmarks compared to physicians, exploring accuracy, limitations, and implications.