New ARC-AGI-3 benchmark test reveals that even the most advanced AI models fail to solve game-like puzzles, scoring below 1% on tasks designed to detect when artificial general intelligence arrives.

ARC-AGI-3 test uses game-like puzzles that require AI systems to solve problems on the fly without prior training

Current top-performing AI models score below 1% on the benchmark, indicating significant gap from human-level general intelligence

Test was created by a research foundation as a potential early warning system to detect when AGI has been achieved

The benchmark focuses on reasoning and problem-solving abilities that distinguish general intelligence from narrow, specialized AI capabilities