🔍 One model writes code, another finds errors in it - and it works.
New Codex on GPT-5.4 has significantly improved in performance. If you are writing your projects on Claude, here’s what you should try: open Codex, switch the model to GPT-5.4, and select the deepest analysis mode. Then give it a simple prompt:
"Check the entire project, analyze it, and find critical errors, as well as what needs to be fixed. Without making any corrections, write to me."
After that, it will think for a while and compile a list of critical errors for you. And already in that same Codex or through your Claude, you can fix all these errors. I assure you, there will be a lot of interesting things to find.