Codex automatically consumed my token reset opportunity during long-running task #28525

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

26.609.71450

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Darwin 23.6.0 arm64 arm

What issue are you seeing?

019ecf0c-6fbd-7560-82ab-45aaf20a333c
019eb08a-b7d5-7392-8737-ad5655cee92c

I am a ChatGPT Pro user and I encountered an issue where Codex automatically consumed my one-time token usage reset opportunity without an explicit confirmation from me.

Around 1:00 PM on June 16, Shanghai time, I started a Codex task and then stepped away. When I returned, I found that my available token usage reset opportunity had already been used automatically. I suspect this may have happened because the 5-hour token usage limit was reached during the task.

However, at that time, my weekly usage allowance still had approximately 80% remaining, so I did not expect Codex to automatically consume my reset opportunity.

In a recent task, GPT-5.5 was unavailable due to capacity issues（Selected model is at capacity. Please try a different model.）, so I switched to GPT-5.4. The task then ran for around two hours. During that run, there were more than ten automatic context compression events, which appear to have consumed a large amount of my 5-hour token usage.

From my perspective, this seems to be caused by product behavior rather than intentional user consumption. The reset opportunity was used automatically, and the excessive token usage appears to have been caused by repeated automatic context compression during a long-running task.

I would like the team to review this case and restore my token reset opportunity and/or adjust the excessive token usage if appropriate.

What steps can reproduce the bug?

1. Start a long-running Codex task using a Pro account.
2. Use GPT-5.4 when GPT-5.5 is unavailable due to capacity.
3. Let the task continue running for around two hours.
4. During the run, Codex performs repeated automatic context compression events.
5. The 5-hour token usage is consumed quickly.
6. If a token usage reset opportunity is available, Codex may automatically consume it without explicit user confirmation.
7. The user returns and finds that the one-time reset opportunity has already been used.

Expected behavior:
Codex should not automatically consume a user's token usage reset opportunity without explicit confirmation, especially when the weekly usage allowance is still largely available.

Actual behavior:
My one-time reset opportunity was consumed automatically during the task, even though my weekly usage allowance still had approximately 80% remaining.

What is the expected behavior?

Codex should not automatically consume a user's one-time token usage reset opportunity without explicit confirmation.

If the 5-hour token usage limit is reached during a long-running task, Codex should pause the task and ask the user whether they want to use the reset opportunity, instead of using it automatically.

Also, automatic context compression should not repeatedly consume excessive token usage in a way that quickly exhausts the 5-hour usage limit. If repeated context compression is required, the user should be warned clearly before a large amount of usage is consumed.

Expected behavior:

1. The reset opportunity should only be used after explicit user confirmation.
2. The user should be notified before the 5-hour token usage limit is exhausted.
3. Repeated automatic context compression should not cause unexpectedly high token consumption.
4. If this happens due to product behavior, the consumed reset opportunity and excessive token usage should be restored or adjusted.

Additional information

I am a ChatGPT Pro user（zylena@icloud.com）[019ecf0c-6fbd-7560-82ab-45aaf20a333c ; 019eb08a-b7d5-7392-8737-ad5655cee92c]. This happened around 1:00 PM on June 16, Shanghai time. My weekly usage allowance still had about 80% remaining, but my one-time reset opportunity was automatically consumed during a long-running Codex task. The task also had more than ten automatic context compression events, which appear to have consumed a large amount of 5-hour token usage. I would like the team to review the case and restore the reset opportunity and/or adjust the excessive token usage if appropriate.