<p>While a vast number of new metaheuristics have been developed to solve truss optimization problems in the past few decades, the prominence of classical gradient-based methods has been ignored. Although the performance of metaheuristics is continuously improved, many of them are still developed for solving such simple truss single-objective optimization problems with less than 30 design variables. It is very questionable whether the development of more and more complex metaheuristics for solving such problems is necessary while classical gradient-based methods are also capable of solving the problems with good performance. For this reason, the primary aim of this study is to prove the outstanding performance of gradient-based optimizers over some state-of-the-art metaheuristics on single-objective truss sizing optimization problems. Five classical gradient-based methods are employed to compete with five state-of-the-art metaheuristics on nine 2D and 3D truss sizing optimization problems. The problems range from small-scale to large-scale problems with up to 942 design variables. Based on the obtained results, the hypothesis of this study has been confirmed. The performance of the metaheuristics falls short when compared to the gradient-based methods, specifically sequential quadratic programming, which emerged as the clear winner in this study.</p>