LLMs can play chess fine.

The best model you can play with is decent for a human - https://github.com/adamkarvonen/chess_gpt_eval

SOTA models can't play it because these companies don't really care about it.