Need to have serious discussion with the $200/month plan

Hi everyone, I’ve been a long time user and supporter of codex since September last year with the $200/month plan. I’ve seen ups and downs but lately I pointed out the major degraded perfromance from GPT 5.5. To tell you how serious this is, I’ve had to push back production releases due to numerous issues that was performed by codex and its failure to adhere to instructions. I’ve never had codex cause this much problems to the point of delaying releases but is happening and we are still unable to get it to rectify and fix the regressions it has caused throughout our codebase. I’m currently relying on Opus. 4.8 to fix all of its mistakes and the reality is frightening. Not only are there serious security lapses caused by its degraded ability to follow instructions, I find that a lot of the sloppiness coincides around the time where the usage sync issue was declared fixed. This means that any work it has done since then now is under scrutiny and Opus 4.8 is finding a ton of critical issues and worse xhigh appears to have refactored the code base in a strange way that does not address the prompt instructions but hallucinated big time writing . I’m unable to any other feedbacks but it is quite worrying since we rely heavily on codex and it has been excellent as of date but we are still seeing the same degradation issues I addressed a week ago and others have pointed out. Also there is the larger issue with GPT Pro. This does not seem like the old Pro model where it has been able to think through a problem for extended period of time. It often responds way too quickly and it makes me think that there is not enough inference being performed. Lastly, the usage issue. Overtime I’ve noticed gradual inflation and reduction of usages. Now that the promotion period is coming to an end as of this month, all of these collective issues is making me rethink my workflows purely centered around codex. I hate to say this but this seems to happen repeatedly with new model releases in the past. There’s a honeymoon phase where we seemingly have maximum and optimal inference then it slowly degrades but i’ve never seen it this bad. As a result I am now using Claude , in particular Opus 4.8 and leaning on it to correct the errors caused by gpt 5.5 xhigh. A month ago, this was unthinkable. 5.5 xhigh got a ton of work done and could be trusted but here we are, we seem to be back in the era of 5.3~5.4 suddenly. Once again, I implore the OpenAI team to take this issue seriously and investigate not just my reports but others raised on here and other social media platforms. I want Codex to succeed and I have benefited from it and want others to do the same continuously. Thank you.