In order to effectively solve the problems of insufficient convergence of the solution set and limited engineering practicality in the multi-objective optimization design of long-span concrete-filled steel tubular (CFST) arch bridges, in this study, the Tusha river super large bridge is taken as an engineering case and an improved NSGA-II algorithm, integrating an adaptive crossover and mutation operator and a dynamic constraint-handling mechanism is proposed to achieve the collaborative optimization of bearing capacity, lightweight design, and economic efficiency. The optimization results show: a 12.2% reduction in the weight of the main arch ring, an increase in the bearing capacity safety factor to 1.83, a 14.5% decrease in the positive bending moment at the midspan of the main girder, and a 17.8% reduction in live load deflection. The overall cost is reduced by 8% compared to the original design. Further analysis of algorithm performance indicates that the convergence and distribution of Pareto solution sets of the improved NSGA-II significantly increased the hypervolume indicator HV increases by 22% compared to the traditional algorithm), and all optimized schemes strictly meet the requirements of design specifications. The findings of this study can provide theoretical support and practical references for the multi-objective optimization design of complex arch bridge structures.