This study employed latent profile analysis (LPA) and latent transition analysis (LTA) to examine the short-term longitudinal development of L2 reading comprehension among young Korean EFL learners in Grades 1 and 2 participating in an AI-integrated gamebased learning program. L2 reading comprehension was assessed at three time points across vocabulary, sentence-level, and discourse-level comprehension. The LPA identified five distinct reading profiles, revealing substantial heterogeneity prior to formal English instruction. The LTA showed pronounced upward transitions during the early phase of the program, particularly among lower-performing learners, followed by strong profile stability in the later phase. These findings suggest that the instructional environment may support initial improvement in foundational reading-related skills. However, the limited later profile transitions indicate possible constraints in sustaining continued developmental progress. The findings highlight both the potential and limitations of AI-supported game-based learning and underscore the importance of examining heterogeneous learner trajectories in early technology-mediated language learning contexts.