Drawing holds significant promise for supporting students’ science learning. While drawing science models enables students to engage deeply with science content, teachers struggle to assess students’ science sketches because it is prohibitively labor-intensive. Recent advances in generative AI models introduce the opportunity to automatically assess students’ science drawings. We present a multimodal LLM-based science sketch assessment framework for automatically analyzing student-drawn science models. First, we collected a dataset of science models sketched by 8–12 year olds. Next, using evidence-centered design, a science model sketch rubric was developed, and the science models were assessed by human experts. Finally, our science sketch assessment framework was utilized to analyze the science models. Empirical evaluations of the assessment framework using two foundational multimodal models were performed. Findings from the evaluation indicate the science sketch assessment framework accurately assesses student science models, paving the way toward automated formative assessment for supporting students’ sketch-based science modeling in the classroom.