I’m not sure honestly. It could be some combination of bad spatial reasoning of the LLMs and lack of any training data for this specific challenge.

You can see replays for all of the matches if you hover over the cells in the table.