New benchmark confirms AI models still perform poorly at visual perception

The Decoder The Decoder —

Four clocks and colorful stacked cubes set against abstract shapes illustrate phases of time and modular process steps.https://the-decoder.com/wp-content/uploads/2026/08/perceptionbench-nano-banana-pro.jpg" style="height: auto; margin-bottom: 10px;" width="1920" />


Moonshot AI's PerceptionBench tests how well multimodal AI models can actually "see," separate from logical reasoning.

No frontier model reaches 60 percent accuracy, and GPT-5.6 Sol leads by a narrow margin.

Many supposed reasoning errors actually happen as early as the image-reading stage.


The article https://the-decoder.com/new-benchmark-confirms-ai-models-still-perform-poorly-at-visual-perception/">New benchmark confirms AI models still perform poorly at visual perception appeared first on https://the-decoder.com">The Decoder.

Read full article at The Decoder →