Generative AI for Code Synthesis: A Comparative Study of Large Language Models in Software Engineering

– This paper presents a comprehensive analysis of large language models (LLMs) in automated code generation, comparing architectural approaches, security implications, and practical efficacy. Through systematic evaluation of GPT-4, GLM-4, CodeGen, and Polycoder models, we identify key performance metrics in code quality, vulnerability rates, and development efficiency. Our findings reveal a 30-100x efficiency gain in AI-assisted programming compared to traditional methods, with GPT-4 achieving 42% vulnerability-free code versus CodeGen’s 39% in Python implementations. The study highlights the critical need for robust validation frameworks when deploying LLMs in software engineering workflows.