ARC-AGI-3

ARC-AGI-3 benchmark introduces a new standard for evaluating AI agents' general intelligence through interactive reasoning and adaptive skill-acquisition in dynamic environments. This represents a fundamental shift from static puzzle-solving evaluation toward testing AI's ability to learn and adapt in interactive, real-world-like scenarios. The benchmark is critical for measuring progress toward true artificial general intelligence beyond narrow task performance.