10 months later, the test still stands. Neither GPT-5 nor Cursor's new stealth "code-supernova" model can solve this (GPT-5 thought for 2m 26s).

Can't add images to reposts, so their new drawings will be in the comment.

(Reposting my Pythagorean tree test: nilmamano.com/posts/my-personal-test-for-do-llms-still-li98)