What is ARC AGI 2?

ARC-AGI-2 is a new benchmark designed to push AI systems to their absolute limits in general problem-solving. It builds on François Chollet’s original ARC (Abstraction and Reasoning Corpus) but introduces a next-level collection of tasks that are trivial for humans yet perplexingly hard for current AI. These tasks span pattern recognition, basic physics, logic puzzles, and other abstract challenges. Pure LLMs score essentially 0% on ARC-AGI-2, highlighting its difficulty. Even advanced multi-modal or neuro-symbolic models struggle, making ARC-AGI-2 a litmus test for progress toward robust general intelligence. The benchmark’s goal is to identify gaps in current AI reasoning and drive research toward models that can learn new concepts with as little information as humans. In essence, ARC-AGI-2 represents “AI’s hardest test,” where success would signify a major leap toward human-like general reasoning (paper).