This study presents a robust and flexible method for text classification using LLMs as its main component. Modern information transfer has undergone transformation through semantic communication which uses large language models (LLMs) to enrich communication channels. The research investigates how multiple LLMs function and perform when used as tools for semantic communication tasks. These factors known as contextual understanding, robustness and efficiency receive analysis to show their practical use in actual settings. The method can handle many types of text classification jobs. While still producing reliable and adaptive results, our approach streamlines conventional text classification operations by cutting down on the amount of pre-processing and deep domain knowledge required. On four separate datasets, we tested several LLMs, ML methods, and neural network models. When compared to more conventional methods, the outcomes show that some LLMs perform better on tasks including sentiment analysis, multi-label classification, and spam SMS identification. Furthermore, our findings show that few-shot learning or fine-tuning can further improve system performance, with fine-tuned models achieving the best results on all datasets. Thanks to their superior capacity to grasp linguistic subtleties, large language models (LLMs) have shaken up natural language processing (NLP) text classification. Pre-trained on large datasets, models like GPT and BERT capture intricate word-to-word correlations, enabling them to excel in tasks such as sentiment analysis, topic classification, spam detection, among others. Transfer learning enables LLMs to train on a diverse range of domains and languages with minimal task-specific data. Their superior performance and accuracy outperform conventional machine learning and rule-based methods, thanks to their capacity to deal with sarcasm, ambiguity, and complicated language structures.