The purpose of this research was to determine the effect of the flipped-classroom and flex blended learning models in enhancing the mathematical problem-solving ability of junior high school students. The quasi-experimental pre- and post-test method was used to carry out this research. The sample consisted of 128 students divided into two equal groups (n1 = n2 = 64). Self-efficacy data were collected through a questionnaire, while problem-solving ability was evaluated using validated mathematics problem test-sets. Analysis of covariance (ANCOVA) was used to analyze the data, with the independent variables comprising the learning model (flipped and flex) and self-efficacy (low and high). The dependent was the post-problem-solving ability score, and the pre-test was the covariate. The test results showed that participants in the flipped class group obtained a final problem-solving ability score greater than those in the flex group after the initial score was controlled (p < 0.001), with a large effect size of 2 = 0.382. Although self-efficacy was a significant factor in the final test score (p = 0.001, 2 = 0.134), the interaction with learning models was insignificant (p = 0.226). This shows that students will increase their math problem-solving ability test scores in flipped and flex classes regardless of their self-efficacy level. In conclusion, the flipped classroom technique can be implemented to enhance mathematical problem-solving abilities among Grade 8 students with low or high self-efficacy. To ensure a successful learning process, variances in cognitive capacity, learning medium, objectives, and students’ emotional qualities also need to be considered.