Imagine a future where AI not only grades your written answers but also understands the diagrams and illustrations you include. That future is closer than you think. Researchers are tackling the challenge of "multimodal" short answer grading, where AI systems evaluate answers containing both text and images. This is a significant step up from traditional automated grading systems that primarily focus on text. Why is this important? Because incorporating visuals often demonstrates a deeper understanding of a subject, especially in fields like science and engineering. Students can express their knowledge more comprehensively through diagrams, charts, and other visual aids. This research introduces a new dataset called MMSAF (Multimodal Short Answer Grading with Feedback) containing over 2,000 examples of short answer questions from high school-level physics, chemistry, and biology, along with reference answers and synthetically generated student responses – both text and images. They tested several leading multimodal AI models, including ChatGPT, Gemini, Pixtral, and Molmo, on their ability to grade these multimodal answers and provide helpful feedback. The results are promising, with some models demonstrating impressive accuracy in assessing the correctness of answers and the relevance of included images. Pixtral, in particular, stood out for its alignment with human judgment, generating feedback that was more nuanced and insightful. However, the journey isn’t over. There are still challenges to address, like ensuring fairness and accuracy in grading diverse visual representations and scaling the creation of these multimodal datasets. But this research lays the groundwork for a future where AI can provide richer, more personalized feedback on student work, ultimately fostering a deeper understanding of complex subjects.