Collective responses to localized perturbations are essential for the adaptability of animal groups. Using a biologically grounded computational model of burst-and-coast swimming in Hemigrammus rhodostomus, we investigate how group size and the strength of social interactions shape collective dynamics under perturbation. The model integrates experimentally derived attraction and alignment rules, behavioral heterogeneity, and boundary effects within a circular tank. We identify four collective states (schooling, milling, turning, and swarming) and characterize a critical regime in which groups exhibit multistable dynamics. At this critical point, small subsets of perturbing individuals, defined by altered social interaction strengths, can induce sharp transitions to new collective states. In particular, such transitions occur in large groups (N = 100) but not in smaller ones (N = 25 or 50), highlighting a size-dependent sensitivity to disturbance. We show that both the nature of the perturbing individuals and the initial state of the group modulate the system\'s responsiveness. Our findings suggest that large groups may exploit criticality to remain both robust, flexible, and that individual variability can serve as a catalyst for adaptive reconfiguration. This work provides new insights into how the internal group structure and perturbation design influence collective behavior in animal groups.