Fig

Frontier reasoning agents from Anthropic, Moonshot AI, and Alibaba failed on interactive 2D mazes, solving only 6 out of 150 runs across 50 mazes, with 45 mazes unsolved by any model. The evaluation, …

Researchers at Fig and Manifold Research Group introduced GUI-Perturbed, a benchmark that stress-tests browser-use AI models by applying controlled UI perturbations like zoom changes. They found that …