Critics challenge ARC-AGI 3 benchmark methodology

Community analysis argues that ARC-AGI 3 restricts reasoning agents from maintaining context across actions, potentially skewing performance results. The critique suggests this limitation prevents models from retaining previously solved steps during evaluation.