You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Codex keeps running after weekly limit is reached, without consuming remaining credits #18018
I reached my weekly Codex limit, but Codex continues to run. The weekly quota is exhausted, and although I still have credits available, they are not being used. This seems inconsistent with the expected billing or usage behavior.
What steps can reproduce the bug?
Use Codex until the weekly limit reaches 0% remaining.
Start a new thread after the weekly limit has been exhausted.
Observe that Codex still runs instead of stopping or consuming the remaining credits.
What is the expected behavior?
Once the weekly limit has been reached, Codex should clearly notify the user that the weekly quota has been exhausted and indicate when it will reset. If credits are still available, Codex should continue running and consume those credits as documented. Otherwise, it should stop running tasks.
Additional information
I initially thought this might be related to a credit booster I apparently received, but I am not sure whether that is affecting the behavior.
Although it keeps running, it is apparently slower than before, because it is taking too much time in thinking in basic tasks.
Same thing happened to me. My 5 hour weekly limit are exhausted mid task but still it kept going without stopping. It would be worthwhile to check if the model is being degraded when the limits are exhausted instead of hard stopping the task (like what happens in the claude code and antigravity)? Or is this a genuinely a bug?
This is by design. OpenAI did this to provide a better user experience (at the expense of their profit margin), preventing the tasks from getting aborted midway. No extra charge happens. Consider this a curtesy from OpenAI and use it with good faith (don't abuse too much).
If will eventually stop though, e.g. if you try to send another message or it tries to compact the context.
This is by design. OpenAI did this to provide a better user experience (at the expense of their profit margin), preventing the tasks from getting aborted midway. No extra charge happens. Consider this a curtesy from OpenAI and use it with good faith (don't abuse too much).
If will eventually stop though, e.g. if you try to send another message or it tries to compact the context.
They were at first thinking as a Product where the User experience mattered but now I guess they don't. If a task is running, and your daily limit has expired but your weekly limit is left, I think the ideal path should be to let the task finish and deduct from the weekly limit for that particular task to finish at least. This should be allowed for the users to pursue a goal or finish a task gracefully. If you are designing for user experience, then this is the bare minimum.
No, it keeps working. I guess I am lucky that I have not received that update yet, but given the codex ample limits in $20 plan compared to claude code, cant complain even if it stops immediately after limits are exhausted.
When i posted it, it was when they made the stricter limit following for free tier. They accidentally applied it to all paid plans. Has since been resolved, I dont have any other issues
I have noticed that it continues for a while until it stops in the free when the task is ongoing and go tier but immediately stops in a prolite plan. But yes, I agree compared to claude it does give more limits which is more practical in my experience.
Been running 5 agents gpt 5.5 xhigh fast mode working on massive project of millions of lines of code. One agent is itself managing a swarm of agents and has called over 600 agents in 24 hours. I hit usage limits for the week yesterday morning and the current weekly usage bar is not even showing anymore above the 5h that is stuck at 100%. Im on pro 5x and downgraded to plus and currently cli says im on plus. Even when red warning saying ive hit limit appears, the agents continue working even when re prompted.
I wont complain, I do have 0 credit for more usage. For the first time I really started to work with "Goals" and when the weekly limit arrive, i got a popup to "add credits" but it is now still working for the last hour and I wont stop it overnight hoping that it just continues :D
Activity
github-actions commentedon Apr 15, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
namevinu commentedon Apr 18, 2026
Same thing happened to me. My 5 hour weekly limit are exhausted mid task but still it kept going without stopping. It would be worthwhile to check if the model is being degraded when the limits are exhausted instead of hard stopping the task (like what happens in the claude code and antigravity)? Or is this a genuinely a bug?
I was using gpt 5.4 with reasoning = high
brunolemos commentedon May 15, 2026
This is by design. OpenAI did this to provide a better user experience (at the expense of their profit margin), preventing the tasks from getting aborted midway. No extra charge happens. Consider this a curtesy from OpenAI and use it with good faith (don't abuse too much).
If will eventually stop though, e.g. if you try to send another message or it tries to compact the context.
Source:
rafaself commentedon May 16, 2026
This is great, but it was not exactly the situation in the issue. Also, it seems to be fixed already.
rafaself commentedon May 16, 2026
This issue seems to be already solved since it is not happening anymore.
siahisaforker commentedon May 28, 2026
they patched this. it immediately stops now. kinda ruined it
thisistuhinsen commentedon May 30, 2026
They were at first thinking as a Product where the User experience mattered but now I guess they don't. If a task is running, and your daily limit has expired but your weekly limit is left, I think the ideal path should be to let the task finish and deduct from the weekly limit for that particular task to finish at least. This should be allowed for the users to pursue a goal or finish a task gracefully. If you are designing for user experience, then this is the bare minimum.
namevinu commentedon Jun 1, 2026
No, it keeps working. I guess I am lucky that I have not received that update yet, but given the codex ample limits in $20 plan compared to claude code, cant complain even if it stops immediately after limits are exhausted.
siahisaforker commentedon Jun 1, 2026
When i posted it, it was when they made the stricter limit following for free tier. They accidentally applied it to all paid plans. Has since been resolved, I dont have any other issues
thisistuhinsen commentedon Jun 1, 2026
I have noticed that it continues for a while until it stops in the free when the task is ongoing and go tier but immediately stops in a prolite plan. But yes, I agree compared to claude it does give more limits which is more practical in my experience.
Lucid-IDE commentedon Jun 3, 2026
Been running 5 agents gpt 5.5 xhigh fast mode working on massive project of millions of lines of code. One agent is itself managing a swarm of agents and has called over 600 agents in 24 hours. I hit usage limits for the week yesterday morning and the current weekly usage bar is not even showing anymore above the 5h that is stuck at 100%. Im on pro 5x and downgraded to plus and currently cli says im on plus. Even when red warning saying ive hit limit appears, the agents continue working even when re prompted.
TheRabbitHoleDev commentedon Jun 8, 2026
I wont complain, I do have 0 credit for more usage. For the first time I really started to work with "Goals" and when the weekly limit arrive, i got a popup to "add credits" but it is now still working for the last hour and I wont stop it overnight hoping that it just continues :D
li01223123 commentedon Jun 13, 2026
it's over
kudasaixc commentedon Jul 12, 2026
Keeps working exactly like that for me.
DrescoAV commentedon Jul 12, 2026
which version of codex are you using? desktopp app, extension or cli? and on what OS? what type of account?
bennypea commentedon Aug 17, 2026
Snitch
DrescoAV commentedon Aug 17, 2026
At that time OpenAI openly shared that's how it worked. Not its not anymore