The increasing focus on computational thinking and numeracy has established critical thinking as a paramount skill in modern mathematics education; however, evidence on how emerging technologies, specifically Augmented Reality (AR), mediate this cognitive process in students remains limited. This study addresses this lacuna by providing an in-depth, mixed-methods analysis of junior high school students' critical thinking performance within an AR-assisted numeracy learning environment, focusing specifically on the challenging domain of rectangular prisms. Employing a descriptive quantitative and qualitative design, this research aims to meticulously describe and evaluate the critical thinking ability of three eighth-grade students selected via purposive sampling across high, moderate, and low initial mathematical ability levels. Data were systematically collected through validated open-ended written tests and follow-up interviews and analyzed using Facione’s framework (interpretation, analysis, evaluation, and inference) with a three-tier performance categorization (> 75: high; 60-74: moderate; < 60: low). The analysis revealed a moderate overall level of critical thinking ability (mean score: 66.62). A direct correlation with prior ability was observed: the high-ability student performed highly across all indicators, the moderate-ability student showed proficiency in interpretation and analysis but deficits in evaluation and inference, and the low-ability student remained low across all four facets. These findings significantly highlight the differential influence of prior mathematical ability on higher-order thinking skills, suggesting that while AR effectively facilitates the visualization and analysis components of critical thinking, complementary pedagogical interventions are necessary to enhance complex skills such as mathematical inference.