Retrieval-Augmented Claude Opus 4.7 and GPT-5.5 Surpass Human Performance on the Nuclear Cardiology Board Preparation Exam (and Claude Drafts a Paper About it)

The authors aimed to determine whether next-generation large language models (LLMs) with retrieval-augmented generation (RAG) could outperform human fellows-in-training on the American Society of Nuclear Cardiology Board Preparation Exam. Their results showed that both Claude Opus 4.7 and GPT-5.5 achieved mean accuracies of over 86%, surpassing the average human score of 78% and indicating the potential of these models as effective educational tools in nuclear cardiology.