With the growing complexity and interdependence of urban systems, multi-objective optimization (MOO) has become a critical tool for smart-city planning, sustainability, and real-time decision-making. This article presents a systematic literature review (SLR) of 117 peer-reviewed studies published between 2015 and 2025, assessing the evolution, classification, and performance of MOO techniques in smart-city contexts. Existing algorithms are organised into four families—bio-inspired, mathematical theory-driven, physics-inspired, and machine-learning-enhanced—and benchmarked for computational efficiency, scalability, and scenario suitability across six urban domains: infrastructure, energy, transportation, Internet of Things (IoT)/cloud systems, agriculture, and water management. While established methods such as Non-dominated Sorting Genetic Algorithm II (NSGA-II) and Multiobjective Evolutionary Algorithm based on Decomposition (MOED/D) remain prevalent, hybrid frameworks that couple deep learning with evolutionary search display superior adaptability in high-dimensional, dynamic environments. Persistent challenges include limited cross-domain generalisability, inadequate uncertainty handling, and low interpretability of artificial intelligence (AI)-assisted models. Twelve research gaps are synthesised—from privacy-preserving optimisation and sustainable trade-off resolution to integration with digital twins, large language models, and neuromorphic computing—and a roadmap towards scalable, interpretable, and resilient optimisation frameworks is outlined. Finally, a ready-to-use benchmarking toolkit and a deployment-oriented algorithm-selection matrix are provided to guide researchers, engineers, and policy-makers in real-world smart-city applications. This review targets interdisciplinary researchers, optimisation developers, and smart-city practitioners seeking to apply or advance MOO techniques in complex urban systems.