State-of-the-art vision-language models fail at interpreting real scientific artifacts that domain experts find straightforward, highlighting a critical gap between general image understanding and specialized scientific visual reasoning needed for practical biotech applications.
VIALS is a benchmark of 161 visual question-answering tasks based on real scientific images (gel blots, microscopy, flow cytometry plots, etc.) from biotech workflows. Current vision-language models struggle with these domain-specific images despite excelling at natural images, revealing gaps in scientific reasoning that limit their usefulness in professional life sciences research.