Repository navigation
Hitting rate limits #6172
Description
I'm aware this isn't the right place to file this issue, BUT it's now clear that a huge cut in limits is being observed. Reddit is talking about it, and your GitHub is full of similar issues. This is being systematically observed by many independent users. We have reached a point where we can look around and confirm it - we are not imagining things. And we observe silence from your end.
Is it the new standard among AI companies to serve a downgraded product and cut limits after a few month?
Thread update: ae-openai clarified that limits weren't reduced. However, a cache issue was identified which caused cache misses and huge usage spikes, with the expected fix shipped today (Nov 4).
Update 2: Today, Nov 5 usage was reset.
Activity
github-actions commentedon Nov 3, 2025
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
ae-openai commentedon Nov 4, 2025
Hey,
lunamidori5 commentedon Nov 4, 2025
Thank you! On top of that I hope those poorly formatted messages that crash the full cloud website also do not count... Thank y'all for your hard work!
iSuslov commentedon Nov 4, 2025
Hey @ae-openai, this is my particular case:
It says limits will reset on Nov 9th, thus it started on Nov 2th and by using it for one day (two 5 hour windows) I've got 80% of my weekly limit reached. You guys should announce this caching issue publicly and reset weekly limits of those who was affected.
techjoec commentedon Nov 4, 2025
Reposting some data here as #6008 (comment) (@etraut-openai ) noted this thread is an "update". Issue I put data into on Saturday post "fix": #6008
Summary
Plan: PRO
Duration to weekly limit hit: Under 24 hours!
Usage "week" started: 11/2/2025 12:10:00 AM UTC.
My week
I ran my numbers for CLI and cloud from the start of my "week" til I was prevented from using the service and this is what I came up with. Might be off a slight hair but likely very close to actuals..
Cloud Totals
CloudTasks: 109
Tokens: 3891767
Total Task Hours: 20.25
Cloud Task Averages
Tokens per task: 35704
Minutes per task: 11
CLI Session Totals
InputCachedTokens - 352834688
InputTokens - 375819446
OutputTokens - 2073269
ReasoningTokens - 1042304
Five-hour usage window sample
As for 5 hour usage...The below activity (and the CLI is only end of session stats, and some were likely started before that current 5-hour window reset so likely a bit inflated!
Cloud Totals
CloudTasks: 29
Tokens: 1075448
Total Task Hours: 5.38
Cloud Task Averages
Tokens per task: 37084
Minutes per task: 11
CLI Session Totals
InputCachedTokens - 90468224
InputTokens - 96415999
OutputTokens - 653123
ReasoningTokens - 351552
Serena MCP is the only extra tool I use, and Codex CLI makes sure to forget about it for 99% of the session anyway.
Overall 5 hour span of usage
Daily usage from OpenAI directly
Unsure if https://chatgpt.com/backend-api/wham/usage/credit-usage-events reports "date" in UTC or not...But if it does, then the first day can be fully ignored. But it doesn't even add up to 100%...Confusing.
{
"date": "2025-11-01",
"product_surface_usage_values": {
"cli": 4.734244660872603,
"web": 19.14151781164927,
"github_turn": 0.06114338122898196,
"github_code_review": 0.7460948304726966
}
}
Soooooooooooooo
@etraut-openai This is "working as intended"? Please advise ASAP.
The site says:
Pro Usage: Usage limits apply across both local and cloud tasks. Average users can send 300-1,500 local messages every 5 hours or 50-400 cloud tasks every 5 hours, with a shared weekly limit.
That is really not the case here it seems.
msbealo commentedon Nov 4, 2025
I'm not a fancy CLI user, but I have been using the web interface for some projects. However, on Thursday, the 30th, I was hit by this issue. I was suddenly rate-limited, significantly below my previous usage. I struggled on for a few days using a local environment in VS Code, but it didn't help much. My 5-hour limit was being reached after 2-3 prompts. Now my weekly limit has been reached, and I have to wait 5 more days!
In the meantime, I've been investigating alternatives, but this has significantly affected my progress.
pontes-a7 commentedon Nov 4, 2025
any update? @ae-openai
techjoec commentedon Nov 4, 2025
The only update I've seen is that they will stop charging us for when Cloud fails. That should be a no-brainer and should have already been in place before billing us, but that aside there's something seriously wrong OR the dozens/hundreds of us complaining (see OpenAI's discussion boards, reddit.com/r/codex/, here, etc) are either stupid or crazy (tell us which, please). I've spent (wasted) hours providing numbers and many others have too. OAI seems to feel this is working as intended apparently.
These comments here are frustrating: https://old.reddit.com/user/embirico especially "web tasks are longer". Uhhh..I average 11 minutes overall. ~100 cloud tasks and ~22mil non-cached tokens from CLI ate 100% of my Pro weekly limit!?!?
As for CLI usage, this is a recent dump I did of the token usage. I haven't really changed my habits at all, in fact have done things to try to streamline, but apply_patch and sed taking up such a gigantic share of the token load, and often on the same damn files over and over, surely is not helping. It feels like after making such a gigantic usage-impacting change with sudden jerk-of-the-wheel motions that slow-rolling tiny "improvements" now really isn't helping one bit here.
I hate being a whiner, but at $200/month having my entire weekly quota blown in under 24 hours is absolutely bonkers and not aligned with the stated plan limits from what I can tell. And my usage is hobby strictly, OAI is banned and CIDR blocked at work for Reasons(tm). I feel for the folks getting jammed with this that rely on it for their day jobs!
Also this one is gold right here: #6198 Poor guy had codex change one line and lost 8% of his Plus budget. Yeah, tell us how that is "working correctly" please??
techjoec commentedon Nov 4, 2025
And to add, the messaging from OAI is borderline sloppy. Saw a few tweets today now that "in the next two weeks" things might improve. Then nuggets like this:
So which is it, team? Are us users stupid/crazy, or is the platform broken??
techjoec commentedon Nov 5, 2025
Ah well. Canceled my Pro. Guess that's all I can do. Best of luck to all and I hope they fix things. :(
The interesting part of it is that I tried to get the "AI Support" to open a ticket for me before doing this, but it insisted that my usage is advanced and suggested I pay more for credits. After a half dozen prompts I said "Ok cancel my Pro" and without any hesitation it fully canceled my plan. No confirmations or anything...ha! I was going to keep Plus for a month, but the kick in the butt was clear!
aitikcho commentedon Nov 5, 2025
It was a good cloud for coding, however ... today I have tried 3 tasks, all of them failed :
By the time of these 3 unsuccessful tasks I am 80% of my 5 hour limit and 25% of my 5 day limit. I used to code with this tool and now it seems I will have to use another tool. Hope I get good suggestions.
I’m sorry, but I’m not able to complete this request.
why?
Copy
I’m not sure why the earlier response said the request couldn’t be completed—there’s nothing in the project setup that prevents working on it. If you’d still like to proceed, let me know and I’ll be happy to help implement the project-specific data handling and dropdown.
msbealo commentedon Nov 5, 2025
It looks like my usage was reset this evening. I don't know if this has been applied to others, or if it's related to a support ticket I put in a couple of days ago, or what. While I appreciate the reset, I also notice that ONE prompt has used up 17% of my 5-hour usage. It's nice to be back. I certainly haven't got Kilo Code working as well as Codex does on my project, but I hope that this isn't just a temporary fix.
46 remaining items
pfahlr commentedon Nov 17, 2025
I did go ahead and try claude code, and in quite a few ways it works better, especially enjoying the plugins and ease of adding MCP servers... but the one way it is not better... unfortunately... is this one... wow I burned through my allotment quickly. I just noticed I had 250$ in claude code for web (they've apparently just added a web/cloud environment coding version similar to codex for web) and it expires at midnight nov 18... I think I'll be able to use it all up as I've managed to burn through $60 in credits over the past couple hours.
It could be the AI companies have decided to work together to price fix... and since all these companies have been able to replace 5-10 engineers with a single engineer running one of these tools, what's 200$ a day when it replaces 5-10 engineers that you had to pay 300-500$/day? It's still a bargain.
Unfortunately, for many a developer... it's not a bargain for us.
winnal commentedon Nov 19, 2025
From the latest update, the usage limits issue seems to be fixed. Usage is going down way slower now. Even when using the new max model on xhigh effort.
neo-picasso-2112 commentedon Nov 20, 2025
super disappointing. I am actually considering leaving this platform altogether and going open-sourced. What a disappointing rug pull from the OpenAI and Codex team.
rymatoi commentedon Nov 21, 2025
@etraut-openai any updates? I don't believe you marked it as a bug 3 weeks ago and it still persists 🤡
rymatoi commentedon Nov 21, 2025
@etraut-openai if you can't fix it you can hire me, I'll do it for you
techjoec commentedon Nov 22, 2025
Asked support very nicely and clearly three times now. Fully ignored for going on 7 days and @etraut-openai and others avoiding this thread completely. Jumping on other minor issues though and linking other issues to this black hole consistently however. Ah well bank charge back and burned some big customers for good. 👋🏼
etraut-openai commentedon Nov 22, 2025
We posted an update on this topic in reddit and X a couple of days ago. Sharing here as well.
Update on Codex usage
Over the past weeks, we’ve been working to increase usage limits and fix bugs. Here’s a summary of progress:
Usage increases since Nov 1
Fixes & improvements
Measuring the improvements
That’s a lot of improvements and fixes! Time to measure the lifts—unfortunately we can’t just look at the daily usage data powering the in-product usage graphs. Due to the multiple rate limit resets as well as changes to the usage limits system to enable credits and increased Plus limits, that daily usage data in the past is not directly comparable.
So instead we verified how much usage people are getting by looking at production data from this past Monday & Tuesday:
Bear in mind that these numbers do not reflect the expected 30% efficiency gain from GPT-5.1-Codex-Max, which launched yesterday (Nov 19th). We expect these numbers to improve significantly more!
Summary
Codex usage should now be more stable and higher than it was a month ago. Thanks to everyone who helped point out issues—we’ve been investigating them as they come and will continue to do so.
techjoec commentedon Nov 22, 2025
But tough luck for those of us that didn't get the credits and no comment on support refusing to acknowledge it?
etraut-openai commentedon Nov 23, 2025
Hi @techjoec. Sorry, but I don't have any information about your particular case or your interactions with support. My understanding is that credits were given to users as thanks for their patience as we investigated the issue. Reading through your posts above, it sounds like you had already canceled your subscription by that point.
aaronedev commentedon Nov 23, 2025
i think my limits are back to normal level, can anyone confirm the same? I just tested and ran prompts + multiple agents etc. Seems it stabilized itself
techjoec commentedon Nov 23, 2025
Oh I did, after begging you all for review. I posted extremely detailed logs, examples, and even burned my credits generating them. Full out ignored until after the 'offer' expired. Did you see how I canceled?? I am sure you are busy (no /sarcasm at all) so maybe not but I tried repeatedly to reach out to support via the dang support tool, and it refused to give me a way to make a ticket. So I literally said the below and it just did it....I was actually expecting it to give up on fluff and actually create a support ticket, and I would have stayed Plus at least. This was on Nov 4th, merely hours before you announced on reddit the credit.
The chat interaction after begging for a support ticket with the support "agent" that canceled my sub Nov 5, 2025, 5:06 AM UTC:
The specific wording you posted about the credits does not disqualify me. On the Pro plan I used Codex web during the period you stated:
I was patient. I posted many times to this issue AND others including verbose stats and log outputs. I begged for clarity and communication as it was impacting us. I got nothing in return. Here we are over two weeks later, and I am being told that the credits were as a thank you for being patient and thus I have no claim to them? I feel I was very patient and eager to help resolve the issue for everyone, tried to provide you and the codex team with data to consider, and was just flat ignored.
To this date support has ignored my last three emails requesting escalation or even a case number and explicitly denied your claims of any credit offer even exists.
Support:
:(
etraut-openai commentedon Nov 25, 2025
I'm going to close this issue. There are still a few issues discussed above that we are tracking, but they are covered by other issues:
#2330
#6674
#5999