This research presents and evaluates an automated assignment assessment framework using GPT-5. The approach addresses limitations of manual grading by providing a scalable, consistent, and objective alternative for assessing student work across diverse academic disciplines. Educators can upload student submissions, optionally accompanied by grading rubrics, which the framework uses to produce evaluations based on either the uploaded rubrics or general academic standards. Key features include cross-document consistency checks, bias mitigation to support fair and responsible grading practices, and the generation of detailed formative feedback. The methodology was evaluated using a dataset of 604 authentic student submissions from eight university-level STEM courses—including Data Science Fundamentals, Probability Theory and Mathematical Statistics, Econometrics and other. A comparative analysis between human-assigned grades and model-generated evaluations revealed a strong Pearson correlation coefficient of 0.9569, with the majority of grade discrepancies falling within one grade point. These findings demonstrate the potential of large language models to enable consistent, scalable grading aligned with the principles of responsible AI and established academic standards.