Revolutionizing Scientific Figure Decoding: Benchmarking LLM Data Extraction Performance

Large Language Models have demonstrated impressive capabilities in natural language understanding and processing. However, as AI and LLMs continue to evolve, their ability to accurately and efficiently interpret data from scientific figures and plots remains obscure. In this study, we test and evaluate the ability of state-of-the-art large language models (LLMs) in analyzing and extracting information from scientific figures commonly found in materials science literature, including stress-strain curves, heatmaps, 3D plots, contour plots, and other visual representations. We utilize a benchmark dataset comprising figures derived from several published works to evaluate the model's performance in quantitative material property extraction. Preliminary results highlight both the potential and current limitations of LLMs in handling visual scientific content, pointing toward future opportunities for AI-assisted data extraction in materials informatics.