<p>This study investigates the state-like versus trait-like nature of self-efficacy among 1,767 Chinese eighth-graders in the context of mathematical problem-posing and problem-solving. The investigation utilized problem-posing and problem-solving tasks developed by Cai et al., alongside task-specific self-efficacy instruments based on Bandura’s social cognitive theory. Using both Linear Regression and Generalized Additive Models, the analysis revealed that the correlation between self-efficacy and performance was consistently stronger for problem-solving than for problem-posing. A key finding was that task-specific self-efficacy was not always the strongest predictor. Specifically, self-efficacy for posing easy problems emerged as the most robust predictor for all levels of problem-posing performance. Conversely, self-efficacy for solving easy problems was the strongest predictor only for easy and moderate problem-solving tasks. These results suggest that self-efficacy exhibits a “State-Trait Duality,” simultaneously functioning as a temporary state shaped by situational factors and a stable trait consistent across tasks. This dualistic framework provides a more nuanced understanding of how self-efficacy influences students’ mathematical engagement and achievement.</p>