In recent years, metaheuristic algorithms have evolved into powerful and flexible optimization tools capable of addressing complex engineering problems. This study presents a comprehensive comparative evaluation of six well-established metaheuristic algorithms, including Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE), Grey Wolf Optimizer (GWO), Whale Optimization Algorithm (WOA), and Harris Hawks Optimizer (HHO), for optimizing the lateral load pattern of the pushover analysis in reinforced concrete (RC) moment-resisting frames. The main objective is to assess and compare their computational efficiency, convergence behavior, and prediction accuracy within a unified and consistent optimization framework. Four three-dimensional RC buildings with 2, 4, 6, and 8 stories were designed in compliance with the Iranian Seismic Design Standard No. 2800 and the Ninth National Building Code. The central bay of each structure was extracted and modeled as a two-dimensional frame in the OpenSees platform to perform nonlinear analyses. Each model was subjected to nonlinear time-history analyses using forty real earthquake records categorized into four sets of ten records. The mean interstory displacements obtained from these dynamic analyses were defined as the target displacement profiles for the optimization process, while the mean roof displacement was adopted as the control parameter for the pushover analyses. During the optimization process, the lateral load pattern of the pushover analysis was iteratively updated to minimize the mean squared error (MSE) between the pushover and target displacement profiles. All optimization algorithms were implemented in Python, with GA and PSO executed using the PyGAD and PySwarms libraries, and other algorithms developed through customized coding. Communication between the optimization scripts and OpenSees was established via TCL-based scripting and text file exchange. Each algorithm was executed for fifty generations with a population size of twenty, resulting in more than ninety-two thousand pushover analyses in total. The results revealed that computational time significantly increased with building height and that algorithm performance should be evaluated simultaneously in terms of runtime efficiency, convergence stability, and displacement accuracy. Optimization of the lateral load pattern of the pushover analysis substantially reduced the discrepancies between pushover and nonlinear dynamic results, particularly for low-rise frames where relative errors were typically within a few millimeters. The computational cost differences among algorithms expanded with height; for example, DE required approximately 140% more runtime than GA in the two-story frame, while this gap exceeded 400% in the six-story frame. Rapid error reduction in early iterations did not necessarily ensure optimal final accuracy, and a stable convergence trend in later stages proved to be a more reliable indicator of algorithmic robustness. For low-rise frames, all algorithms produced acceptable results, with runtime and convergence stability serving as the main differentiators. In mid- and high-rise frames, performance variations became more pronounced, and both the earthquake record set and frame height exerted a decisive influence on the optimization outcomes. Under pulse-like records, the algorithm choice was more sensitive to record characteristics, where WOA generally performed best in short-period (SP) sets, while GWO and occasionally PSO yielded superior performance in medium- and long-period (MP and LP) cases. DE consistently delivered the highest accuracy but with a very high computational cost, GA and PSO provided faster and well-balanced solutions, WOA exhibited smooth and stable convergence, and HHO demonstrated the weakest performance overall. Collectively, these findings provide a clear understanding of the behavior and efficiency of metaheuristic algorithms in optimizing the lateral load pattern of the pushover analysis and offer a foundation for future development of machine learning models to predict and generalize optimal load patterns for RC frames with different characteristics, thereby enhancing the accuracy and reliability of seismic performance assessments.