<p>This paper introduces the Mandrill Optimization Algorithm (MOA), a novel multi-swarm and derivative-free metaheuristic inspired by the dominance hierarchy and foraging behaviors in mandrill hordes. The main contributions include: (1) a multi-swarm framework that enhances exploration in high-dimensional spaces; (2) distinct separation of exploration and exploitation phases, dynamically balanced via elapsed time ratios; (3) memory-efficient design requiring no historical data storage; and (4) innovative mathematical models simulating mandrill movements for improved convergence. MOA addresses slow convergence and local optima trapping, particularly in complex optimization problems. Evaluated on 40 mathematical test functions (fixed-dimensional, multimodal, composite, and high-dimensional unimodal), MOA demonstrated superior performance: on multimodal benchmarks, it achieved an average 35% lower fitness value and 20% faster convergence compared to PSO and ABC (based on mean results over 30 runs, with p &lt; 0.05 via Wilcoxon test). On high-dimensional functions, results were competitive, matching or exceeding GSA and SHADE in 75% of cases. In real-world applications across five engineering problems (e.g., tension/compression spring design), MOA reduced optimal costs by up to 15% over traditional methods. These outcomes highlight MOA's robustness, making it a powerful tool for multimodal and high-dimensional optimization.</p>