<p>Since the beginning of the twenty-first century, the flipped classroom, as an innovative teaching model, has garnered widespread attention. However, most previous meta-analyses have focused solely on its effects on learning outcomes within a single domain, and the reported effect sizes have exhibited considerable inconsistency. This study employed a second-order meta-analysis, integrating 50 first-order meta-analyses and yielding 75 effect sizes of the flipped classroom on student learning outcomes. It systematically examined variations in the model’s effectiveness across different types of learning outcomes, disciplines, and instructional contexts. The results showed that the flipped classroom significantly improves students’ academic performance, cognitive ability, and non-cognitive ability. Additionally, the study revealed significant heterogeneity in effect sizes and identified significant differences across disciplines and learning levels. Meta‑regression analysis indicated that the moderating effect of learning level was no longer significant in the multivariate model, whereas discipline remained a significant moderator of the flipped classroom’s effect sizes. The effect sizes for language learning and public health were higher than those for other disciplines, suggesting that discipline may be more fundamental than learning level, and that the differences in effect sizes by learning level may stem from a structural overlap in the distribution of disciplines.</p>