“Codex Rate Limit 2 weeks” - I just noticed. Is there some info on this ? How does this effect the limits in detail ?
Hi everyone,
This topic is for discussing Codex limits after the move from message-based usage to token-based usage.
Additionally the 2x rate limits promotion has ended and some users have reported hitting limits sooner after this change. This is the right place to compare experiences, share what seems to trigger it, and discuss what helps make usage last longer.
Under the new system, bigger prompts, larger repos, long context, large AGENTS.md files, MCP servers, Fast mode, image generation, and model choice can all use up limits faster.
Current Codex usage and pricing details are here:
Seems like I made a silly mistake (fall into trap?) and bought 6 seats of 1-year business plan, just was impressed with Codex capabilities.
After 5h limit change I can do approx 10x-15x less of what I was doing just a week ago.
Made bunch of skills, thought I could automate my business flaw a bit and kaboom!
Today I hit 5h limit within half task which literally feels like ten or more times less of what I was doing. I could understand cutting limit twice or more…but 10 times??
Is this a bug or the company “strategy“? I think I’m not alone here…
Your not alone, its happening to “most” if not “all” of us even on very simple things.
I really wish it’s a bug…
It’s very much real. Our multi-seat account hit 5 hour limits within minutes and our credit pool which used to last about a month, gone in 2 days. I almost made the mistake of switching the business over to yearly subscription but decided to hold off a few days before pulling the trigger. It’s extremely unfortunate and I hope it’s an honest mistake, because otherwise this would be one very dishonest business move on their part. This is very painful now that we’ve wrapped so much of our automation and development around it.
Yeah and my business partner was like “dude buy monthly don’t go for a year“…I was like “come on what can go wrong, we’re going to use it anyways and I kinda want to support amazing team of OpenAI for the great tool“…LOL
I started a not-terribly-ambitious project back in February, creating an in-house tool that improved email filing in Clio.
At first, everything worked great. I was building my project one slice at a time in my free time after work. I was blissfully unaware that usage limits even existed.
Then, around mid-March, I started hitting my weekly usage limit after 3-4 days of use. Same usage pattern: maybe 10-12 prompts on a heavy day, all written by ChatGPT, about 2/3 of which were just small, corrective passes to fix bugs. I am using Codex through its VS Code integration, which is supposed to burn less usage than the web interface, and have been using 5.4-Mini where appropriate to minimize usage even further.
On April 3, after my weekly usage quota had reset, I tried to get started again. This time, I was able to run one feature-hardening pass (not even creating a new feature, just fixing one that already exists) and two tightly-focused passes to fix bugs after the hardening pass. Those three small, focused prompts used up my entire 5-hour quota and 12% of my weekly quota.
This is completely unacceptable. I am trying to build something in my spare time on a Business-tier subscription and cannot run more than three small passes without hitting usage limits. I can’t imagine trying to use Codex for anything more ambitious or full-time. At this rate, it will take me months to complete a project that is already substantially finished.
If your goal was to push everyone away from using Codex for anything, well, congratulations, I guess. Mission accomplished. You’ve created a hammer that can only be swung three times per day. Great job.
Seeing the same on my side — and honestly, this isn’t workable.
A few small, focused passes burning through quota breaks any real development flow, even for simple tasks. That’s especially hard to justify on a paid tier.
I appreciate the promo period — it was genuinely useful.
Rolling back to normal limits is fair.
But this feels like an overcorrection. At this level, Codex becomes hard to rely on for consistent work.
Also worth noting: decisions like this create an opening for alternatives (GLM-5, Kimi K2, etc.) to offer more predictable conditions.
Hope this gets reconsidered.
Yeah, i have also opened another topic regarding this matter, thread: “codex-limits-finally-good-after-april-1st-reset” floating in newest topics.
Its going nuts now. I am running Business accounts with multiple user for multiple devs.
Yesterday and today the workflow is basically this:
- Dev logins into CODEX on IDE
- Sends 3 tasks
- Dev must logout, ask me for new login approval
- Dev then does again 3 tasks, logs out
- To infinity this loop…
Weekly limits seems like fair and ok IMO. But 5H its going absolute nuts. Each task eats 20-30% of 5H range.
You can actually test it by just writing something simple, just talking with it and you gonna see that it will suck 10% to answer 
Also i have noticed the quality of the work went down from codex
But yes, its is becoming absolutely unusable at this point.
Im running GPT 5.4, Medium, Full access, Locally - no MCP connectors, nothing just purely controlled by core agent. So my tasks are super efficient and accurate.
We tired working on GPT5.3-codex, little better situation - you can produce just a little more but still same - 5H gets wiped off super fast.
The only choice is API or new Codex account credits at this point but budget must be super high, you can easily spend 50-100$ a day.
Suggestion: give admins the option to make Codex usage a pooled resource. I work for a law firm with 10 ChatGPT users on a business-tier license. Nobody else in our workspace is using Codex at all. If our account admin could pool our Codex quotas, I wouldn’t constantly hit work stoppages due to usage limits.
While pooling quotas wouldn’t fix the problem for everyone, it would at least fix it for similarly-situated users who burn through their quota while other users in the same workspace never even touch theirs.
Honestly that feels like an insane take. The problem is most certainly the way OpenAI decided to deal with things, like these credits and limits, but Codex, since 5.3, is absolutely blasting away any other LLM I’ve ever seen when it comes to coding.
I COMPLETELY agree.
I hadn’t noticed it yet, possibly because I hadn’t closed my Codex in days… But now I have hit the new limits and it’s exactly as others are saying. I just gave it 3 minor prompts, thinking nothing of it as I rarely hit my 5H limit anyway… And then it happened… 3 minor prompts, and; You have hit your limit!.
WTH?! This is absolutely, 100% unusable.
it’s about on par with Claude code and it’s pro plan usage limits, which has always been the community’s baseline for poor and unusable.
7 prompts used my entire 5 hour limit, in 10 minutes. They weren’t even useful prompts. I don’t know what they are thinking with this.
One prompt just took 67% of my 5 hour usage on another account.
This is codex 5.3
Compacting twice in between the first promptbefore it began, and once after when I tried a second prompt.
At this rate I will get 8 prompts a week per project.
Since the latest changes, introduced in Apr ‘26, our ChatGPT Business seats have become nearly useless. A simple prompt with a well defined unit of work now consumes nearly the entire 5hr limit and the weekly limit is proportionally equally evaporated. As a business which has been using Codex since its inception, this is the first time we’ve had any such issues and hence I’ve resorted to this post.
I primarily noticed when our credits pool, which we would generally only hit lightly during extremely heavy sprints, has now burned through 10k+ credits in just 1 to 2 days, on top of three standard ChatGPT Business seats.
There has been a noticeable change also in the output structure of at least the GPT 5.4 model in the past few days. Not sure if related to the consumption change, but there appears to be much more feedback from the model and I believe much more context utilization for setup before action. I’m seeing in some instances nearly 40% of the 258k context used compared to around 25% for the same codebase and similar tasks.
To keep this succinct - this service has become unaffordable and unsustainable for us overnight.
We’ve had no other significant issues and generally experience good performance from the models working in highly technical projects, spanning many scientific/engineering domains.
At best with the recent changes, i could get a max of 3 prompts in 5 hour period. That translates to 5 hour window finished in less than 20 minutes total time. And includes me making coffee in the kitchen and coming back. So most of the time spent was me entering a prompt and to eventually come back with coffee.
And the quality of responses for the lousy 3 prompts was also lousy. ChapGPT also didn’t finish the task in any capacity. It delivered broken code. Not just logically wrong, but broken too.
I signed up for Claude 5x. Maybe I will upgrade to the 20x. It’s been working great.
Same here, current codex 5h limit is ust UNUSABLE!!
Time to try Claude I guess
You first gave generous limits, then doubled them for 2 months (Feb 2 → Apr 2). People built workflows and projects around that capacity.
Now you’re not just reverting — you’re cutting usage ~8x from what users were actively using.
This is not normalization. It’s a regression.
Impact:
-
breaks trust in platform stability
-
discourages building on Codex
-
signals that capabilities can be arbitrarily taken away
There is also no proper mid-tier ($50 / $100). The choice is: use far less or leave.
Many will leave.
This is not just about limits — it’s about trust.
Current pattern looks like: give a lot → create dependency → sharply restrict
That damages reputation and pushes users to alternatives.
If Codex is meant to be a serious dev platform:
-
limits must be predictable
-
pricing must match real usage tiers
-
changes must be gradual
Right now, this change does the opposite.
Hot take?
You could say: thank you for all the free compute?
Instead of manipulating Ai to make fallacies for you?
”breaks trust in platform stability” = false claim.
”signals that capabilities can be arbitrarily taken away.”
= another false claim.
They literally told you the exact date it was going to happen.
You CAN get 2 accounts, at the 20-dollar rate, and use the desktop app.
You CAN do 3 accounts….
You CAN also use your regular ChatGPT to give you tactical prompts to make the most out of the computer.
My question to you is, after that much free compute, where are your apps making you the money?
That’s what it was for.