In multi-objective symbolic regression, the objective is to improve the model’s accuracy while minimizing its complexity. It results in a Pareto front, including a fair compromise between accuracy and complexity. In this study, we propose a hybrid cooperative genetic programming approach containing the hybridization of NSGA-II and an adaptive weighted multi-population GA, which cooperatively optimize both models’ accuracy and tree length. In the weighted multi-population GA, the weights are assigned adaptively. We also propose a new version of offspring selection to suit the needs of multi-objective symbolic regression. The two algorithms communicate solutions with each other in specific intervals. The proposed algorithm is tested on the Feynman benchmark datasets, and the results are comparable to the NSGA-II in terms of accuracy and models’ tree length.