Is Codex becoming the new Claude Code? DUMB AGENT and MODEL

[YoloMark98]

Hello everyone, I’d like to express a concern about CODEX that I already had to face in the past with Claude Code, namely that, week after week, the model’s performance was becoming increasingly inefficient to use... and I’m noticing this quality degradation in recent days with Codex as well. My question is: are we on the same path that Anthropic took? I thought OpenAI didn’t have such problems, but with the launch of SORA the hardware for extended reasoning models must have been reduced a lot...

Paying $200 a month now means nothing. I’d even pay $500 if it were possible to always have the same excellent performance. I switched from Claude Code to Codex exactly a month ago because of Claude Code’s degradation issues, but just this past week I’ve noticed that CODEX can’t perform at all. I’m worried, but that’s how it goes with these big companies, sooner or later they always fuck you.

[YoloMark98]

Let me know what do you think and if you're facing same issue :)

[freelanceagent1]

no, I use claude at the same time right now, decided to give it a try after not using it since codex released & I can tell you that claude is still claude and codex is still superior :D

[guidedways]

@YoloMark98 I would strongly disagree. Codex is at the moment (in practical sense) the best model / cli (compared to claude code / gemini). I have max plans for all three and regularly compare them with similar prompts but Codex outshines everything else at least for complex problems. Everyone's mileage will vary given every project and usage requirements are different, however Codex is incredibly powerful when it comes to reasoning. The /review feature and its ability to meticulously review changes for regressions is outstanding.

Having said that, I've recently discovered a lifehack with codex: #7296

Using a custom system prompt may help as it replaces the base / default prompt with what you want it to be (or things you would like it to focus on holistically).