This paper benchmarks a diverse range of models - spanning conventional machine learning, deep learning, and large language models (LLMs) - for the task of Vietnamese clickbait detection. While traditional classifiers and sequential architectures offer competitive baselines, transformer-based models and fine-tuned LLMs demonstrate substantial improvements, particularly in handling nuanced and ambiguous headlines. We also examine the role of prompting strategies, highlighting their impact on zero- and few-shot performance. Through standardized evaluation across model families, this paper provides a comprehensive comparative analysis and establishes a unified reference point for future research on clickbait detection in low-resource languages like Vietnamese.