Compute is limited. They will need to sunset some models to move onto new ones. It’s always been this way.
If they were to stick with 5.3 and never release 5.6 they would fall behind the competition.
But nevertheless feedback is clearly useful because it helps them know when they’ve done something right with a model.
precisely. Well put, and fully agree. 
If it were reasonably priced and delivered comparable work, no one would say a word. 5.3 Codex is a highly demanded model with an ideal price-to-performance ratio. It would have been simpler to just remove 5.4, keeping what became important and convenient for the community.
Don’t forget they are also likely reducing subsidies at the same time so they can head to profit.
So long as there is fair competition we should at least respect their search for profit?
The situation is a bit unfortunate. They waved the carrot and now they have to take some of it away. But this is not untypical. I think the industry is just finding its way.
It’s very possible AI agent coding is significantly more expensive to serve than it felt to begin with.
okay now i give up on my OpenAi sub i was subbed for codex since the beginning and the extentionn in VS code Called CODEX but doesn’t have codex ??? that’s crazy to me ?
i never thought i will use Calude code but apperntely OpenAi wants us to leave so we will do so ?
im also trying something called Kilo Code and will see hopefully it’s better than codex
I wanted to share some feedback as a long-time user and paying Pro subscriber.
The primary reason I subscribed was GPT-5.3-Codex. For my workflow, it was the most effective coding model I’ve used, including compared to competing tools. It was fast, focused, predictable, and exceptionally good at understanding existing codebases without excessive token usage.
With the removal of GPT-5.3-Codex, the main value proposition that brought me to the platform is unfortunately gone. While GPT-5.5 is undoubtedly powerful, it feels much more general-purpose and significantly more token-intensive for the kind of software engineering work I do day-to-day.
Unless GPT-5.3-Codex returns or a comparable coding-focused replacement is made available, I will discontinue my Pro subscription and rely on Claude Code for my development workflow going forward.
I share this respectfully and as genuine product feedback. GPT-5.3-Codex was an excellent model (much better than Claude Code in my humble opinion), and I believe many developers subscribed specifically because of the experience it provided.
Thank you to the team for all the work that went into it.
Quick feedback: I agree with all the feedback shared here. I’ve been using codex 5.3 pretty much daily and thought it was impressive. It executed to perfection whatever it was asked to do.
I tried 5.4 yesterday and couldn’t really produce what i wanted. A lot of back and forth with a disappointing result. I then tried codex 5.3 through the API and within 1h consumed almost 8 USD… so that’s really not viable for me.
I’m afraid I’ll be signing up for something else and keep an eye just in case a different codex is brought back.
Open AI vs XAI, who goes bankrupt, and who will bring us closer to the singularity…
This is absolutely absurd. I have a $200 Pro subscription and use Codex 5.3 with VSCode in my IDE to create most of my scripts. If there were a reason, I would cancel my pro subscription and just maintain my Super Grok Heavy subscription at 300, this would be the final straw for me. Grok Build Beta dropped recently, so I will use their TUI instead of Codex 5.3 in my Terminals from now on. This is a grave financial decision for OpenAI. I have put up with a lot regarding OpenAI, their censorship, and their extreme bias toward safety and avoiding lawsuits rather than usefulness and obedience. I have seen XAI pay for the lack of censorship. There was a happy medium, Codex 5.3, was it. OpenAI just showed me that fear rules them. They prefer to cover their ass over usability, so Grok will always win and be superior. If you want to change the world, you can’t be afraid of your tool. XAI just won. I will not be placated by 5.3 codex spark, or GPT 5.4, or GPT 5.5, the sheep believe are upgrades. You can fool most, but you do not fool me.
I’m still somewhat flabberghasted that these guys think it’s ok to do what they’ve done. They got the data. They know 5.3-Codex was our work horse. It’s cool you wanna sell us your Ferrari but we just need a car we can afford to drive man. Wallet votes need to be cast.
OpenAI just silently pulled all Codex models. Now we are left with simple model choice:
That is it. On paper 5.4-mini is better then Codex-5.3. Reality differs completely.
There is already a mile long thread of users complaining about the rug pull. OpenAI will not answer. Frankly, compared to the money grab Anthropic is doing, OpenAI is amost kind.
Excuses aside, I propose we answer in kind. Please stop complaining, simply cancel subscription.
Go to https://openai.com/, login to ChatGPT, click on your profile icon → Settings → Billing and click Cancel subscription. We can all survive nicely on chinese models for a while - they are no more crap than GPT-5.4-mini.
Compute IS limited, however I respectfully disagree. They need to sunset SOME models, but I think you know as well I do that sunsetting incredibly token efficient and high-quality models and claiming “fleet management” and compute limits is a farce. They are literally consuming more compute by forcing everyone onto 5.5/5.4.
Doing it to a Codex model is the biggest load as well “Let’s pull a smaller, more efficient model from a very small and far more technical subset of our users…not because users are refusing to use our flagship though…nope not that!”
This is a cash grab.
Just found out you can access GPT-5.3-Codex and GPT-5.2-Codex via Github Copilot. I guess we all know where we will be subscribing now. But then again, I do not like the Copilot agent, it is not up to par with the Codex agent or Claude Code CLI agent.
Please, OpenAI, please I beg you, put codex models back on the codex cli and vs code plugins. Pretty please.
This is a general trend: gradually increasing the user cost in order to become profitable. Well… OpenAI is not going to say it quite like that, but that is clearly the idea, and there is nothing scandalous about it.
Like all of you, I notice that with 5.4 high, I burn through my Plus quota very quickly, and my needs are soon frozen by the 5-hour limit. It is the same with Claude, isn’t it? And that is how they reached their goal of becoming profitable, isn’t it?
Anyway… We are all early users here, compared with the general population, especially in agentic mode. We benefited from the novelty and the boost it gave us; now the company has to become profitable. And guess what? We will either pay up or leave because we cannot afford it.
C’est la vie ! 
That’s what I did and found out Claude is better anyway
Could you elaborate? In what way is Claude better, and how do you use it, with which tools?
It can run commands directly on an OpenShift cluster — for example, when building a new feature, it can test code live inside the pod to verify it’ll actually work in that environment. Highly recommend giving it a try!
I’ve only been using it for a day, so I can’t do a full Claude vs. Codex breakdown just yet. I still have a Codex subscription through my company, but I’ve switched my personal AI account to Claude. The issue with the company account is that tokens get depleted fast with so many people sharing it, so I needed a reliable alternative to keep working when that happens.
Bring back gpt-5.3-codex! I like this model that much better than 5.4 and 5.5 I think
Just when I thought life was going smooth, Codex blows up and pushes me to spend more money. This is nothing more than a cash grab and I for one will not forget that fact.
School is now out and I will do everything in my power to move our school district away from codex. Time to look around…
