Social crowds refer to human crowds with social behaviors driven by the psychological states of pedestrians. The dynamics of social crowds can be modeled mathematically at different scales, in which the evolving psychological states are modeled by introducing one or several “activity“ variables into the modeling framework. The dynamics of the activity variables can interact with the mechanical variables, thus affecting the interaction rules and the collective behaviors of the social crowds. Models of social crowds are useful for describing heterogeneous behavioral features in crowds, especially the dynamics of pedestrians with evolving psychological states. Several illustration examples are given for each scale. At the microscopic level, an individual-based model incorporating activity as stress or learning level effectively captures V-shape formations. At the microscopic level, a leader-follower model demonstrates that higher learning ability can enhance evacuation efficiency in dense crowds. At the macroscopic level, a generalized hydrodynamic model with activity modeling stress or contagion awareness reveals that high stress can accelerate movement while contagion awareness can promote crowd avoidance during the evacuation. The simulation results highlight the necessity of incorporating activity as a third variable besides position and velocity in crowd models to better understand the heterogeneous behaviors in crowds and optimize evacuation strategies.