Codex weekly usage limit decreases while idle with no active tasks

What version of the Codex App are you using (From “About Codex” dialog)?

Latest Codex App version available as of September 7, 2026

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Windows 11 x64

What issue are you seeing?

I have been noticing strange behavior with the Codex weekly usage limit for about a month. I am on ChatGPT Plus.

I am not talking about the 5-hour limit, but specifically the weekly limit.

Sometimes the weekly percentage changes immediately after starting Codex. For example, I can see 97% remaining, open a chat, and shortly after it shows 96%. This does not happen every time, so at first I assumed I was simply seeing the usage meter update or recalculate.

Today I noticed a clearer case.

I finished a task and left the Codex App open, but there were no active tasks running. Codex showed 10% remaining on my weekly limit.

About 10-15 minutes later I checked again. I had not started any new tasks and nothing was actively running, but the weekly limit had dropped from 10% to 9%.

This makes me wonder whether:

• usage from completed tasks is being reported with a delay;
• Codex continues making background requests after a task has finished;
• opening or keeping a Codex session active triggers additional usage;
• the weekly usage meter is periodically recalculated rather than updated in real time.

From the user side, it looks like the weekly limit can sometimes decrease while Codex is idle and no task is running.

I have noticed similar behavior occasionally for about a month. Today's 10% -> 9% change during 10-15 minutes of inactivity made it particularly noticeable.

It would be useful to know whether this is expected delayed usage accounting or whether Codex may actually be consuming weekly usage while idle.

What steps can reproduce the bug?

1. Open the Codex App and check the remaining weekly usage limit.
2. Run a normal Codex task and wait until it is completely finished.
3. Make sure there are no active or running tasks.
4. Leave the Codex App open and do not send any new prompts.
5. Check the weekly usage limit again after 10-15 minutes.

In my latest case, the weekly limit showed 10% remaining after the task was finished. I left Codex idle with no active tasks. About 10-15 minutes later, it showed 9% remaining.

I have also occasionally seen the weekly limit decrease immediately after opening or resuming Codex, for example from 97% to 96%.

The behavior is intermittent and does not happen every time. I have been noticing it for about a month.

What is the expected behavior?

Once a task has completely finished and there are no active tasks or new prompts, I would expect the weekly usage limit to stop decreasing.

If the change is caused by delayed usage accounting rather than actual background usage, the UI should make this clear so that users can distinguish delayed reporting from new usage.

Additional information

This does not happen consistently, which makes it difficult to reproduce on demand.

I have been noticing similar unexplained changes in the weekly limit for about a month. Sometimes the percentage changes shortly after launching Codex or opening a chat, even before any substantial new work is done.

The clearest observation so far was today: the task had already finished, no new prompts or tasks were started, and Codex remained open and idle. The weekly remaining usage changed from 10% to 9% approximately 10-15 minutes later.

I am reporting this mainly to determine whether this is delayed usage accounting/UI reporting, or whether some background activity can actually consume weekly allowance after a task has completed.

Subscription: ChatGPT Plus.

No source files, tests, or entry points are identified in the report. First reproduce the described idle-period change by recording the weekly percentage before and 10–15 minutes after a completed task, then trace the usage-meter or accounting path; done means determining whether delayed reporting or background activity explains the decrease and covering that behavior with a regression test.