Evaluating AI-Assisted Assessment: A Comparative Study of GPT-4 and GPT-5 in Automated Grading of Student Programming Submissions

The increasing integration of AI-assisted tools in higher education offers new opportunities for supporting the assessment of student programming assignments. This paper examines the use of large language models in automated grading by comparing two generations of AI systems, GPT-4 and GPT-5. During the 2024/25 academic year, student submissions from midterm examinations in an introductory C programming course were evaluated using GPT-4 with rubric-based feedback, alongside manual grading by the instructor. After the release of GPT-5, the same set of submissions was reevaluated under identical conditions, enabling direct comparison. The analysis focuses on the alignment between AI-generated and instructor-assigned scores using quantitative metrics, as well as qualitative differences in feedback and error interpretation. A subset of 30 student solutions was examined in greater detail to illustrate differences in rubric interpretation. The results show observable differences in grading behaviour across model generations. However, due to the limited sample size, findings are interpreted as exploratory and emphasise the role of AI as a support tool rather than a replacement for human assessment.