GraphARC: A Comprehensive Benchmark for Graph-Based Abstract Reasoning

In three linesGraphARC is an AI benchmark for abstract reasoning on graph-structured data, generalizing the ARC paradigm to graph transformations. Current language models fail on full graph transformation tasks despite understanding graph properties, revealing a comprehension-execution gap.

Summary generated by Claude — human-verified