Fable and GPT 5.6 Can't Beat This 3D Maze?

I loooooooove finding very hard benchmark for LLM where they absolutely suck at it and just drag their little silica feet and despair because it's a great teaching moment (for me). knowing where the current systems suck help us figure out how they work and what is still lacking in term of capabilities so that jolly researchers can get to town and fix it up with good ol' ducktape. well today my guys we are going to explore mazebench which is a beautiful and poetic labor of love from patience cave and his brother (cave dweller) where they push the poor models on the brink of insanity like a lo