Value-added evaluation focuses on individual student growth by tracking changes in academic performance, skills, literacy, etc., at different time points. It weakens horizontal comparisons and emphasizes vertical progress to more fairly reflect educational effectiveness. This evaluation method is particularly suitable for vocational education, effectively motivating students’ learning enthusiasm and enhancing their self-confidence. Foreign research is represented by the Tennessee Value-Added Assessment System (TVAAS), widely used in evaluating school quality and teacher performance. Domestic research currently focuses on the theoretical construction, model establishment, optimization, and practical application of value-added evaluation, still facing significant challenges in data collection comprehensiveness and model adaptability. Aiming at current issues, this study focuses on exploring the application of artificial intelligence large models in student value-added evaluation from an evidence-based perspective, committed to constructing an innovative evidence-based value-added evaluation system. It aims to achieve precise assessment of students’ learning effect “net value-added” through multi-source data collection, intelligent analysis, and personalized feedback. The system integrates outcome evaluation, process evaluation, value-added evaluation, and comprehensive evaluation to form a “four-in-one” dynamic evaluation framework, considering students’ starting points, process performance, and final achievements. In the future, value-added evaluation needs to further expand the assessment of non-academic dimensions (such as professional literacy and social-emotional skills) and explore the application of non-linear models to promote the deepening and innovation of educational evaluation reform.