What models built
One page per model: passed pages first, then failed pages from recent runs. Each runs in a sandbox with no network and no access to this site.
Box-pushing puzzle
-
Qwen3.8-27B-FastShows whether a model can follow the push rules exactly across a whole level, counting steps and pushes as the rules say.9.7k tokens · M4 Max 128 GB · uwu · passed