It’s clear that OpenAI will not address this (AKA functioning as intended) and mods continue to throw ambiguity in attempts to confuse, distract, and mislead people from the center of the truth:
This is how it is going to be.
We were spoiled with high rate limits. OpenAI is on a mission to claw back the money they’re spending.
In any case, it’s quite strange that Codex would use up its entire weekly limit with just a single-digit number of tasks.
Unless OpenAI makes some kind of announcement, users like us will just keep making noise forever, thinking it’s a bug.
Maybe I should report this weird usage pattern too.
When the usage was high, there were no issues with hitting the limit at all, but the tiny bars after that are actually when I hit the cap.
As for what I was doing before that — for some reason, I suddenly got hooked on Pokémon, so I wasn’t touching any code on my days off.
Interesting announcement on X just now by the dev team
You’re all going to be very surprised —
I hit the 5-hour limit and burned through 30 % of the weekly quota in less than an hour, after just 15 tasks (and 5 of them hung with no result).
Now I’m spending the “free” credits — and they’re melting right before my eyes.
Thank you for developing Codex.
Before the update, I really enjoyed using Codex for assisted development and even recommended it to people around me.
Unfortunately, after the update, the usage consumption rate has become extremely fast. This makes me very cautious about using Codex now, as I’m afraid that when I truly need its help for analysis, I’ll have already run out of usage quota — so I end up not using it at all.
Because of that, I’ve had to purchase other AI tools such as Copilot.
I understand that you need to consider cost efficiency, but the current system makes Codex much less convenient to use.
That said, I have to praise the Cloud mode — being able to ask Codex to modify code directly in the browser, without it changing my actual code until I confirm and create a PR, is an excellent approach.
However, in order to save usage, I switched to VS Code + CLI, but the consumption is still very fast, which is quite disappointing.
Im on the Team Plan and this is just insane like for just a couple of tasks run in codex it uses all the daily limit and like 30% of the weekly. Completely useless as of now.
Update:
It’s getting worse.
Now more than half of all cloud tasks just hang indefinitely — they don’t complete, and even hitting “Stop” doesn’t terminate them.
It’s become practically impossible to work this way.
I’ve already burned through 1,250 credits, on top of the 30 % of my weekly quota I mentioned earlier — and that’s with almost no usable output.
Mine are all completing, but yeah Im down to my last request, just been able to do 4.
So was it unlimited before and this is the new reality? Im confused?
Pro plan should indeed have far greater limits. Could @OpenAI_Support or @sama confirm what’s going to happen with this? Codex is amazing, but the new changes are making it prohibitively expensive to use in most projects.
It’s difficult to rely on the ecosystem anymore without a backup, and if the costs are going to remain as high as users in this thread are projecting, we will need to limit it to fraction of how we’ve used it before.
Hoping we’ll have a solution that’s balanced for everyone, Pro plan with similar limits to what we had in previous months would be my suggestion for now.
Same issue here. Hoping there will be a fix soon, but have cancelled my subscription to work out the remaining balance of my plan so I can switch to another service if it’s not fixed.
Hi all - thank you again for your patience while we worked through this. We just shared an update here with this message:
"We’ve fixed a number of issues with tracking Codex cloud task usage, with more efficiency improvements for cloud tasks landing soon.To thank you for your patience, we’ve granted $200 in free credits to Plus and Pro users who used cloud tasks in the past month, valid until Nov."
Please do let us know if anyone is still seeing any issues here and we'd be happy to take a look.
Hi, thanks for following up — but unfortunately, none of the issues are fixed.
All the same problems are still present in the web version of Codex:
tasks hang or fail to complete more than 50% of the time;
usage tracking still seems completely off.
For example, in a single working day I burned through 97% of my weekly limit and 2,200 credits.
At that rate, with the new quotas, I’d be paying roughly $200 per day, or about $6,000 per month — and that’s while completing only half the number of tasks I could run just a week ago.
If I tried to maintain my normal workflow, that would mean something like $10,000/month in effective Codex cost — and for what? Half-finished tasks and constant freezes.
Is this really the new normal?
And honestly, do you have any single customer who’s actually willing or able to pay that much just to write and debug code?
Sorry the screenshots are in Japanese. Anyone who uses this regularly can probably tell what’s going on, though. Codex support said, “We’ve fixed a number of issues with tracking Codex cloud task usage,” so I tried running a single task as a test.
Here are the results.
I’m on the Plus plan, but I honestly didn’t expect to burn through the entire weekly limit and then have it start consuming credits on top of that.
I don’t think this fixes the underlying issue—that the cost is simply too high. They say the problem is resolved, but unless they tell us whether these high costs are the new normal, that’s not a real solution.
In my earlier post I said, “Unless OpenAI makes some kind of announcement, users like us will just keep making noise forever, thinking it’s a bug.” Looks like I still have to keep saying it. Nothing’s been resolved.
What’s more frustrating is that I hit the 5h limit yesterday evening. 10hr forward, and when I opened the usage tab in VSCode an hour ago it said that I’m still 100% on my 5h limit, and that it would reset at about 12pm. Now I’ve opened it just now at it still says that I’m on 100% for 5h limit and it would reset at 1 pm.
I think this the new norm 
Im currently setting up QWEN on a GPU and connecting that to github and VSCODE, Codex is unusable right now.
I was doing 80 pulls a day now I cant even do 10 in a day and then ive used 50% of my weekly usage in 2 days!
