Impact of Large Multimodal Language Models (MLMs) with Screen-vision in Higher Education: A Quasi-Experimental Study

Abstract As screen-vision multimodal AI tools increasingly enter higher-education learning workflows, empirical evidence about their pedagogical impact—and potential risks for teaching and learning processes—remains scarce. This study examines the potential impact of Large Multimodal Language Models (MLMs) with screen-vision capabilities on students’ performance and perceptions during controlled learning activities among master’s and undergraduate students at University of Alicante. A quasi-experimental mixed-methods design was employed. Two cohorts were established: a control group using Gemini (an MLM without screen-vision) and an experimental group using Google AI Studio (with screen-vision). Perceptions were assessed via the Technology Acceptance Model (TAM) questionnaire, alongside pre- and post-tests, assessment rubrics, and a focus group. Results provide evidence of task-dependent benefits associated with screen-vision, with the clearest between-group differences observed in organization-related activities. Students generally evaluated Google AI Studio positively, although technical limitations and ethical concerns were also reported. The study discusses implications for instructional practice and offers recommendations for integrating screen-vision AI in higher education.