In this paper, we evaluate the performance of ChatGPT and DeepSeek, large-scale language models (LLMs) for natural language processing (NLP) tasks, across five major NLP tasks: sentiment analysis, topic classification, text summarization, machine translation, and text implicitness. We use a structured experimental protocol to ensure fairness and minimize variability by evaluating both models on two benchmark datasets per task using the same neutral prompts. Our experiments show that DeepSeek outperforms in classification stability and logical reasoning, while ChatGPT outperforms in tasks requiring fine-grained understanding and flexibility. These results provide valuable insights for selecting the appropriate LLM based on task requirements.
