@techreport {pub6200,
	title = {Benchmarking LLM for Code Smells Detection: OpenAI GPT-4.0 vs DeepSeek-V3},
	author = {Ahmed Sadik AND Siddhata Govind},
	year = {2025},
	month = {December},
	abstract = {Determining which Large Language Model (LLM) is superior for code smell detection is a complex challenge. This study aims to establish a systematic methodology and evaluation matrix to address this question. We introduce a curated dataset containing smelly code implementations of identical scenarios across four major programming languages: Java, Python, JavaScript, and C++. Each dataset entry is annotated with known code smells, serving as ground truth for evaluation. We benchmark two state-of-the-art LLMs{\textemdash}DeepSeek-V3 and OpenAI GPT-4.0{\textemdash}analyzing their detection capabilities using precision, recall, and F1-score. Our evaluation
spans three levels of granularity: overall model performance, performance per code smell category, and detailed performance per individual code smell type. Additionally, we assess the cost-effectiveness of each model by considering their differing detection techniques{\textemdash}token-based analysis in OpenAI GPT-4.0 versus pattern matching in DeepSeek-V3. Furthermore, we compare LLM-based detection with conventional static analysis tools like SonarQube. The findings provide valuable insights for practitioners aiming to select the most efficient and cost-effective LLM for automated code smell detection.},
	publisher = {arXiv},
	booktitle = {arXiv}
}