A new framework for evaluating generative AI models reveals significant disparities in cognitive abilities, with strong performance in verbal tasks but poor results in visual reasoning. This uneven evolution suggests that current scaling methods may not adequately address fundamental limitations in achieving human-like general intelligence.