Figurative language detection, specifically hyperbole and metaphor, presents an obstacle for natural language processing (NLP) in adding complexity to language. The current research investigates the utility of pre-trained language models (PLMs) in this regard, focusing on a set of diverse general-purpose models (mT5-Small, mT5-Base, mT5-Large) and a model trained for the task of figurative language detection (MMFLD). The results show a trade-off between the benefit of having a larger model able to capture more complex figurative language and, as a consequence, achieve scores of 82% accuracy in the hyperbole classification task and 76.26% accuracy in metaphor classification, and the accuracy of the model trained specifically for this task (MMFLD), which produced comparable scores in the metaphor classification specifically. We conclude that both factors, model scale, and task relevance, are key drivers of improved performance in figurative language classification and understanding for low resource language models.