Teaching-Learning-Based Optimization (TLBO) has recently gained recognition as a reliable, accurate, and robust optimization technique for global optimization in continuous spaces. In this study, we propose an Improved TLBO named as ITLBO, that incorporates a novel term in the teacher phase to minimize population variance, thereby enhancing convergence speed and solution accuracy. The algorithm’s performance is evaluated using the CEC2019 benchmark suite, which includes unimodal, multimodal, and composite optimization problems. The results demonstrate that ITLBO consistently outperforms state-of-the-art algorithms, including TLBO, Aquila Optimizer (AO), Whale Optimization Algorithms (WOA), and Salp Swarm Algorithm (SSA), in terms of mean optimization cost and standard deviation. This highlights ITLBO’s ability to find global or near-global optima with improved robustness and reliability. These findings affirm the potential of ITLBO as an effective tool for solving complex optimization problems.