Which is why they brought it up as something they are trying to improve.
replyLess than other frontier models. Which is scary honestly.
replyNo. GPT models follow instructions significantly better than Claude models.
replyYou tell it too research a repo to find a piece of code it will. Claude will just read the README and guess.
I have a codex session I am using to vibe code a db thats being going for like 3 month. Still doing OK. Try that in CC.
replyWhat's the token usage at?
reply