The Arc-AGI-2 is a new AI benchmark introduced by the ARC Prize Foundation that focuses on measuring AI's generalization capability on unseen tasks, particularly at test-time reasoning. The benchmark emphasizes improvement over previous versions by increasing evaluation tasks from ARC-AGI-1's 2019 iteration and requiring deep reasoning instead of mere intuition. The performance metric indicates that current large language models struggle significantly, scoring near 0%, while specialized systems fare slightly better but still under 4%. The accompanying ARC Prize 2025 offers a $1 million incentive aiming to solve the existing challenges in achieving AGI. Past competitions saw notable participation, enhancing research output with numerous papers published. The community expresses varying opinions on the merits of the benchmarks and their impact on fostering breakthroughs in AGI, debating whether they properly incentivize advancement in general intelligence or merely operationalize specific tasks. The Kaggle competition is set to commence shortly, encouraging individual and collective innovation.