Informal comparison
Rubik’s Cube Solving Without Tools
Can a language model solve a scrambled cube given only its text representation?
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The October 27, 2026 results snapshot covers 39 models. Each saw the same positions in the same order, with scrambles of 3, 6, 10 and 20 moves.
Methodology
A simulator checks the proposed move sequence. A model needs one solve to advance; the three-move stage is an unranked qualifier. Each later stage uses ten cubes, so small changes in rank can reflect run-to-run variation.