GPT 5.2 is remarkably good.

- Does better (not good yet) with Wodehouse test (setup, quips, but 5k words!)

- Very "not lazy", and happy to do work

- Very very good at coding - understands the intent and truly gives it a try on CLI. Can let it rip.

- Much better at images, though from a low bar

- Pro is amazing, first model I can just say "critique" and it just does it, including "this is the red team pass" section

- Exceptional economic reasoning - correctly thuoght through the implications of my coasean bargaining paper, and gave some plausible microfoundations

- The model is very good at following instructions step by step (read the repo, then the docs, then the links, then the attached pdf, then opine)

- Github loading still breaks occasionally though, very annoying

- People claimed it is too slow but I didn't feel it, esp not in codex cli