I am frustrated from the recent updates in the Codex. I was previously working with 5.4 and 5.3 models with fast speed and I never ran out of token but after the 5.5 updates.
The weekly qouta started running out after just a small bug fix for Pro if I use 5.4 or 5.5. if I dont use them and use 5.3-codex then the model remain stuck in planning phase for 3-4 hours.
Hardly solves any porblem anymore. Open AI is trying to monopolise and this is not good. A lot of my friends stopped using Open AI because of this behaviour and such updates from the Open AI.
I am attaching the screenshot as a proof where codex only fixed a small chat mode selector modal in six hours and before fixing it, it took 3 hours to plan to it. Instead of vibe coding seems like the platform is intentionally rejecting the customers to either stop using or have atleast 10 pro plans to actually work with something useful.
I thought 5.5 could help but a simple api integration took it 10-12 hours and none of the apiâs were integrated. After it hits context compaction it literally forgets the requirements. it dont even try to read the last chat messages.
I am really frustrated by these updates from the Codex & Open AI and now unfrotunately in order to solve anything i have to zip the entire mono repo and then ask the ChatGPT Pro to help and then asking kimi or Claude to fix those changes and fill the gaps.
I kindly urge Codex and Open AI to look at the developer experience. its degrading fast.
P.S I am considering moving to Open Source LLMs now as having a Mac Studio to run it seems more cheaper then a controlled monopolised market.
You are welcome to discuss Codex issues on the forum, however the official place to report and track them is the GitHub issue tracker for OpenAI Codex.
Many users are reporting such on this forum and in the issues.
A search of the Codex isses using
is:issue state:open label:rate-limits 5.5
If any issue matches your problem, consider adding a reaction on the GitHub issue page itself, not just on the forum topic, as reactions on GitHub help the developers gauge impact and prioritize work.
Make sure to read the issues as sometimes they contain valuable user or OpenAI staff feedback.
If none are exactly the same issue, try searching the existing GitHub issues for a closer match. If you cannot find one, then open a new issue with as much detail as possible.
Codex rate limits for Pro users are currently doubled until the end of the month. If you are seeing rapid and massive usage limit drain, that appears to be a bug.
You can also use the /feedback command to send a report directly through the app. I think that is the easier path for most users to reach the team, while participating on GitHub connects you directly with other affected users and the developers.
Hello Eric the assumption there is that the user is not a FOSS developer, and is not working in a high-security environment (banking for example) where logging in with personal details is a threat-surface.
By excluding FOSS developers you exclude the people working on the bottom of the stack on whom you and the company you work for critically rely.
Yes: I am one such person.
See obligatory XKCD 2347. Google it as this forum prohibits links.
I ran out an entire 6 hour quota in about 12 minutes flat with only 8 messages sent. I havenât bothered since, as using the smaller 5.4 model has frequently caused significant damage.
Yeah, vbâs guidance here is on point. If youâre seeing unusually fast Codex quota drain, /feedback from Codex is helpful because it sends app-side context to the team.
These may also help explain how Codex usage is counted:
If your usage still looks wrong after checking this, please open a support request at support@openai.com and please supply exact time stamp when you face the limit and any details regarding your usage that would be useful. Thank you!
Very interesting I described almost the same problem as you. Iâm a relatively new user, so I canât compare Codex with its previous versions, but Iâm facing the same issue: the tokens disappear before you even really start working. It feels like they overestimated the 5.5 models and are now charging almost triple for using them. I would laugh if it werenât so frustrating.
I understand the sarcasm about my âbank account,â but that is not the issue. The problem is not whether I can afford it. The problem is that Codex limits are unclear, unpredictable, and in real usage the quota can disappear before meaningful work even starts. Also, API pricing and practical Codex usage are not the same thing. Even if GPT-5.5 is âonlyâ around twice as expensive per token and supposedly uses fewer tokens, that does not solve the user-side problem: the product still feels unreliable when the limit is consumed too quickly. My point is simple: if Codex is sold as a serious coding tool for paid users, then access should be more predictable. At minimum, the UI should explain the limits honestly, and there should be some fallback/basic model for audits, tests, or smaller tasks after the main quota is exhausted.
This honestly hadnât occurred to me until you pointed out the problem, but anyone dealing with a standard IT sysadmin will recognise this: Q: âwhy didnât you tell me that in the first place??â A: âbecause you didnât ask?â
Next time I use codex-cli I will ask it the question, âif i use 5.5 for this task, will you run out of âtokensâ before we are even done?â
Letâs just pray the answer isnât:
A: âiâm sorry, youâve reached your token limit by asking that questionâ
Which⌠yeah. There should be clear costs with estimates up-front.
Iâve noticed similar issues lately, especially with longer projects and context handling. The planning phase feels much slower now and sometimes the model loses track of earlier requirements after compaction. Hopefully they improve optimization and developer experience soon because workflow consistency matters more than just adding new model versions.
For those who are still struggling I figured out the solution which is easy and seems like limitless. Credits never run out.
Create a custom GPT and map via actions to your system.
ChatGPT power tbh surpasses codex and response time is few seconds. In less than 10 minutes ChatGPT Heavy Mode fixed everything and while nerfed codex couldnât do it for 5 weeks.
Here is it for everyone who want their lovely Codex power back:
use it through custom gpt actions.
If itâs written by OpenAI where is the credit, inherited copyright notice and licence and why is it under an independent account?
What was the original repo?
Donât mean to sweat you but these things are important and actually against GitHub terms of service.
Something can be considered a Terms of Use exploit if its reason to exist is to overcome a cap of some kind, no matter if it does not actually break the security of the API itself.
Nonetheless it is quite a cool demo of how you can turn a CustomGPT into a coding agent âŚ
reaction on the GitHub issue page itself, not just on the forum topic, as reactions on GitHub help the developers gauge impact and prioritize work.
I described almost the same problem as you. Iâm a relatively new user, so I canât compare Codex with its previous versions, but Iâm facing the same issue: the tokens disappear before you even really start working. It feels like they overestimated the 5.5 models and are now charging almost triple for using them. I would laugh if it werenât so frustrating.
