[2603.24621] ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence

Computer Science > Artificial Intelligence arXiv:2603.24621 (cs) [Submitted on 24 Mar 2026] Title:ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence Authors:ARC Prize Foundation View a PDF of the paper titled ARC-AGI-3: A New Challenge for Frontier Agentic Intelligence, by ARC Prize Foundation View PDF HTML (experimental) Abstract:We introduce ARC-AGI-3, an interactive benchmark for studying agentic intelligence through novel, abstract, turn-based environments in which agents must explore, infer goals, build internal models of environment dynamics, and plan effective action sequences without explicit instructions. Like its predecessors ARC-AGI-1 and 2, ARC-AGI-3 focuses entirely on evaluating fluid adaptive efficiency on novel tasks, while avoiding language and external knowledge. ARC-AGI-3 environments only leverage Core Knowledge priors and are difficulty-calibrated via extensive testing with human test-takers. Our testing shows humans can solve 100% of the environments, in contrast to frontier AI systems which, as of March 2026, score below 1%. In this paper, we present the benchmark design, its efficiency-based scoring framework grounded in human action baselines, and the methodology used to construct, validate, and calibrate the environments. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2603.24621 [cs.AI]   (or arXiv:2603.24621v1 [cs.AI] for this version)   https://doi.org/10.48550/arXiv.2603.24621 Focus to learn more arXiv-issued DOI via DataCit...