Researchers have proposed a new framework for evaluating large language models' (LLMs) ability to support personalized learning. The framework focuses on how LLMs behave in authentic learning scenarios, rather than just measuring their performance on benchmark tasks. In the study, multiple LLMs were tested using a dataset of student responses to data structures questions, and evaluated on their ability to diagnose understanding, generate personalized guidance, and provide actionable feedback. The results show that different LLMs exhibit distinct pedagogical behaviors and that scenario-based evaluation can provide meaningful insights into model behavior in AI-enhanced education. --- Why it matters: This study matters because it highlights the need for more nuanced evaluations of large language models' ability to support personalized learning. By examining how LLMs behave in authentic learning scenarios, educators and researchers can better understand their strengths and weaknesses, and develop more effective strategies for using these models in educational settings. Source: https://arxiv.org/abs/2509.05346