With so many people actively using 5.3-codex, why is it being retired? Why is support for Codex being discontinued? For my standard coding workflow—writing code—version 5.3 is perfectly capable of supporting a full day of development. In fact, I specifically downgraded from version 5.4 back to 5.3; the code generated by 5.4 simply didn’t meet my requirements, and it consumed my usage quota at an alarming rate—I couldn’t even get a full day’s work out of it. We currently subscribe to the Business plan, which covers a large number of users; however, if version 5.3 cannot be restored, I will have no choice but to switch to a different AI service provider.
I got a free month of chatgpt while trying to cancel, so I’ll fully cancel once it runs out, unless a new version of Codex comes along in the meantime that changes my mind. Until then, I’ll probably burn through the chatgpt 5.5 tokens out of sheer protest.
I’ve switched to Claude for now. It feels a bit slower, but with a solid claude.md file it’s more than capable. I’ve also got the free version of copilot as a backup, and it’s actually pretty decent for simpler tasks.
Bring back 5.3-codex, much better for coding than 5.4 and 5.5.
Compartilho da mesma opnião que você, com codex obtive melhores resultados e menos erros em programação. Achei que era impressão minha, mas achei o codex 5.3 melhor até que o 5.5 que gasta muito e entrega pouco.
I need to test this further, but as far as I can remember Codex installed with Personality set in Pragmatic mode. I am using the default settings and don’t have any Custom instructions set either. I noticed yesterday that somehow my Personality setting had changed to Friendly. I moved this back to Pragmatic and 5.4 and 5.5 Low seem to be more compliant. The biggest difference that I notice between 5.3 Codex and 5.4/5 is that the newer models seem to make assumptions and kinda take the prompt with a grain of salt, like it knows better, It then reasons itself into taking an action that is counter to the original prompt. For example, using 5.3 I have been able to take screenshots of working pages, enabling it to notice issues and correct them itself. However 5.4/5 seems to have required major prompting to get its environment to a point where it can take screenshots reliably. One issue came up and I explained that a large amount of white space was in the screens, please check the screenshot and the model immediately assumed the code was wrong and started editing the code to “fix” the space. I have never seen 5.3 make this type of blunder, not even once.
EDIT: I don’t think the Personality setting impacts the models reasoning.. might all be in my head 
please bring back chatgpt 5.3 for codex. i just switched from claude to chatgpt because of 5.3 is more capable and token efficient. if it continues like this, i might have to start trying deepseek or other chinese models 
What should we have to do ? Your 5.4 is not understanding and making the whole useless. I to switch to an older and Trusted Model 5.3 Codex.
So they kept the 5.3 Codex model and didn’t remove it — but only for the $100 plan. The problem is that we liked it, so they figured they could make it more expensive and force people to pay more. It’s available on the Pro plan. Don’t people with lower incomes deserve access?
its on pro plan? i dont see that listed anywhere except 5.3 spark. But I thought that was also incorrect. If it is still on Pro Plan then that is epic to hear. Will just bump up to that. Just curious though, can you confirm it is still there on the Pro plan?
Above is a screenshot from a user showing that the Pro plan lists GPT 5.3. This morning when I logged into Codex, it was in the list, but without access on the Plus plan. After it closed and reopened, it disappeared. I think that on the more expensive plan, everything is available. To know for sure, we would need someone who has that subscription.
ah ok. yeah when I look at it too I dont see it mentioned at all. But if it is confirmed that it is available will be stoked to upgrade.
It is still available in Cursor so am working with it there a bit on a needed basis.
Somehow I had the feeling this day would come.
5.3 codex always nailed whatever technical bug claude was not able to fix, and that justified a 20$ subscription.
I haven’t tried 5.4 or 5.5 yet, which are anyway much more expensive.
Nobody here is buying the fleet reduction bs.
I’ve been giving 5.4 high a try to replace my building of terraform code and an Azure Devops pipeline. Even with explicit instructions in my prompt saying “no literal subscription IDs, tenant IDs, pool resource IDs or agent names in code” - it goes and hard codes them in. I have completely lost trust in using this as a coding agent.
…
and when I tell it to correct it in my next prompt and tell it is has violated a hard constraint, it’s response is:
```
I’m re-checking the kickoff and my changes against it line by line now, then I’ll remove the literal managed-pool name from code and bring the result back into strict compliance with your interpretation of the constraint.
```
“…with your interpretation of the constraint”. Yes, it was my constraint, I said it 
.
I’m going local hosted LLM.
Today I tried to do a basic task of writing a script for monitoring in Zabbix on a 5.4 mini model - the script was written and it took about 30 minutes to finish it.
In the end, I switched to 5.5 to complete the task.
There have never been any problems with codex 5.3, please return it to the plus subscription