Controlled rocking steel braced frames have been introduced as low-damage seismic systems in which controlled base uplift, post-tensioned cables, and replaceable energy-dissipating fuses provide recentering and energy dissipation mechanisms. However, the combined effects of building height and soil–structure interaction on the fragility and collapse safety margin of these frames have not been sufficiently investigated. In this study, the probabilistic collapse performance of controlled rocking steel braced frames was evaluated by explicitly considering soil–structure interaction. For this purpose, 3-, 6-, and 9-story frames were modeled in OpenSeesPy under two support conditions: fixed-base and soil–structure interaction. The soil–foundation system was simulated using nonlinear Winkler springs, and the models were subjected to 44 horizontal components of the FEMA P695 far-field ground motion set. Incremental dynamic analysis curves, fragility curves, median collapse capacity, collapse margin ratio, and adjusted collapse margin ratio were then extracted and compared. The results showed that the median damage capacity decreased with increasing height. In the fixed-base models, the 〖Sa〗_50 of the 9-story frame was 70, 61, and 57 percent lower than that of the 3-story frame at DS1, DS2, and DS3, respectively. Soil–structure interaction caused a limited increase in 〖Sa〗_50 for the 3-story frame, whereas it reduced 〖Sa〗_50 by up to 13 and 12 percent in the 6- and 9-story frames, respectively. All models satisfied the FEMA P695 collapse acceptance criterion; however, in the 9-story frame, CMR decreased by 14 percent and ACMR decreased from 2.48 to 2.13. Therefore, the fixed-base assumption may overestimate the damage capacity and collapse safety margin of taller controlled rocking frames.