Fell into a big trap again~· I always thought codex-gpt5.2-xhigh was the strongest for coding, so I’ve been working it like a horse lately. Only today did I realize it’s not good at writing data analysis code. For example, something that could be done with pandas’ apply function, it handles with line-by-line for loops, which is several times slower. In several cases like this, it was opus 4.5 that corrected it. No wonder it took so long to run this batch of data back then. Looking at it now, each AI really has its own strengths. I really hope the Han Li of the AI world shows up soon.