<p>This study examined the extent to which the Artificial Intelligence (AI)-based tutoring system Edmentum Exact Path influenced Grade 12 students’ mathematics achievement, long-term retention, engagement, and perceptions of AI-supported learning in Iranian secondary schools. A quasi-experimental mixed-methods design was employed with 97 male mathematics students from four public high schools in Iran. Because the study was conducted in regular school settings, individual random assignment was not feasible; therefore, students remained in their intact classes, and the classes were allocated to experimental and control conditions. The experimental group received mathematics instruction supported by Edmentum Exact Path, whereas the control group received conventional teacher-led instruction during the same academic term. Mathematics pretests, immediate posttests, delayed posttests, and engagement questionnaires were administered to both groups, and baseline equivalence was examined before the intervention. The quantitative findings showed no statistically significant pre-intervention differences between the groups in mathematics achievement or engagement. After the intervention, however, the experimental group significantly outperformed the control group on immediate mathematics achievement, delayed mathematics retention, and cognitive, behavioral, and emotional engagement. The qualitative findings from the experimental group’s open-ended survey responses indicated that students perceived Edmentum Exact Path as accessible, effective, motivating, and useful for personalized practice, immediate feedback, error correction, and sustained engagement with mathematical tasks. These findings suggest that AI-based tutoring systems can strengthen mathematics learning when they are integrated as supplementary tools within teacher-led instruction, although the results should be generalized cautiously because the sample was limited to male Grade 12 students from four public high schools in one national context.</p>