New benchmark confirms AI models still perform poorly at visual perception

The PerceptionBench evaluation reveals that leading multimodal AI models currently struggle with visual perception, with no top-tier system achieving 60 percent accuracy. This indicates that many failures previously attributed to logical reasoning errors are actually rooted in the models' inability to correctly process visual input. These findings suggest that developers may need to prioritize image-reading capabilities rather than just language processing to improve the practical performance of multimodal AI.