GraphARC is a new benchmark designed to evaluate abstract reasoning on graph-structured data, addressing limitations in existing benchmarks. It highlights the challenges faced by state-of-the-art language models in performing complex graph transformations, indicating a need for improved graph foundation models.