Abstract: Long-span bridges often serve as critical infrastructures. It is therefore important to accurately assess the immediate structural performance and build high-fidelity finite element (FE) models of these bridges to improve the disaster prevention capacity of urban transportation. Achieving the above goals relies on efficient and reliable model updating and analysis techniques suitable for complex structures. However, most of the previous studies were based on commercial software platforms, which were expensive, slowly updated and had complicated application programming interfaces, and thus limited the development of further in-depth research. Based on the open-source software OpenSees and Python, a program framework suitable for the model updating of complex structures was proposed. An efficient parallel optimization algorithm for FE model updating (FEMU) of complex civil structures was developed in Python. The program interfaces were developed to link the model analysis module and the parallel computing module to realize the parallel analysis and FEMU of complex structures. On this basis, the damage identification of a simply-supported beam was taken as an example to validate the feasibility and accuracy of the program framework. Further, the scaled physical model of Sutong cable-stayed bridge in a shake table test was selected as the object. The modal assurance criterion (MAC) was used to match the simulated modal shapes and frequencies with those identified from the shake table test to form the objective function for the FEMU. The FEMU of the bridge was carried out on a high-performance computing platform to validate the computational efficiency of the program framework. The results show that the framework can effectively tune the high-fidelity FE model of the long-span cable-stayed bridge to match up with the experimental measurement. The error between the simulated frequencies and the physical model was below 1%. Furthermore, the accuracy of the updated model was verified by comparing the measured responses with the predicted ones using the updated FE models under the seismic input with a PGA of 0.1 g. The research outcome can provide references for high-fidelity and data-driven modeling of complex long-span bridges based on open-source platforms.