CodeEduBench: A Multidimensional Benchmark for Evaluating AI-Driven Code Generation Models in C Programming Education

Large language models (LLMs) are a core technology that can facilitate change in programming education. However, the educational applications and impact of LLMs have not yet been thoroughly evaluated. In this research, we develop the CodeEduBench three-dimensional evaluation framework, which assesses the applicability of LLMs for C programming instruction across three aspects: test case creation, code synthesis, and bug fixing. Five widely-used open-source models are chosen for comparative study. The results of this study show that Qwen2.5-coder-32b has excellent overall performance, although its shortcomings in code generation accuracy and error correction. In addition, this study proposes model selection strategies and optimization pathways to provide methodological support for the scientific application of LLM in programming education.