I haven't been fully keeping up, but anyone else notice the same with Cursor? Could be because I use Claude models (GPT-5 was too slow and iffy results). Funnily enough, Codex has been really impressive recently and I've swapped almost completely to using Codex.