The accuracy of numerical substructures is important in online hybrid tests for assessing structural seismic performance. The recently developed online model updating (OMU) technique can improve the accuracy of numerical substructures by identifying and updating parameters using measurements from physically tested substructures. However, the updated mechanical model often depends on the loading pattern, particularly for models with coupled behaviors in multiple directions. Ignoring this influence may result in a large discrepancy between the simulation results and real performance. In this paper, an OMU method with multiple inputs and considering coupled boundary conditions is proposed to identify the parameters of a complex mechanical model for laminated rubber bearings, where the horizontal performance is significantly influenced by the vertical force. In this method, the horizontal and vertical responses serve as inputs for an unscented Kalman filter to identify the parameters of the mechanical model, and the vertical response provides additional information and constraints. Realistic boundary conditions are achieved by the vertical force-displacement switching control and horizontal displacement control. One of the isolators of a six-story steel moment frame was chosen as the physical substructure. A sophisticated mechanical model was selected to simulate the remaining isolators in the numerical substructure, and the parameters were updated online. The results of the hybrid test indicate that the proposed method provides higher accuracy for estimating the post-yield stiffness ratio than the traditional method. The coefficient of variation was 50% lower, and the convergence efficiency was almost five times higher, demonstrating the effectiveness and accuracy of the proposed method.