Summary This mixed-method study investigated how task complexity (low vs. high) and learning mode (individual vs. collaborative) influence students’ performance, time-on-task, and cognitive load when working with virtual patients (VPs). Using a 2×2 factorial design, 75 undergraduate students were randomly assigned to one of four training conditions. In the assessment phase, participants completed both low- and high-complexity VP cases individually. Additionally, twelve participants were interviewed to explore their perceptions of the VP learning experience. While quantitative results did not show significant group differences in performance scores or cognitive load, the analysis of time-on-task revealed that collaborative learning benefited students in high-complex tasks, whereas individual learning was more effective for low-complex tasks. These findings support an adaptive instructional strategy that combines collaborative and individual learning based on task complexity. Dataset The dataset includes both quantitative and qualitative data from participants. The first file, ‘quantitative data’, contains anonymized participant data for performance scores, time-on-task, and cognitive load across training and assessment phases, categorized by task complexity and learning mode. The second file, ‘interview codes’, presents excerpts from semi-structured interviews with thematic codes reflecting participants’ perceptions of collaborative vs. individual learning and task complexity. This dataset supports analysis of how different instructional designs influence learning outcomes in simulation-based training with virtual patients.