Token consumption without use

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #25510
• #25483
• #26756
• #26755

Powered by Codex Action

English translation:

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

26.602.40724

What subscription do you have?

pkus

What platform is your computer?

windows

What issue are you seeing?

Subject: Possible Codex issue: token usage counter decreasing without active use

Hello OpenAI Support Team,

I would like to report a possible issue with the Codex token/usage counter.

I noticed that my 5-hour token/usage quota is decreasing even though I am not actively using Codex. I have not sent any chat messages, I have not executed any model tasks, I do not have scheduled tasks, and I do not have any active automations.

Observed behavior:

• I opened Codex.
• Without using the chat or starting any task, the 5-hour token/usage indicator dropped to 97%.
• Later, still without using the system, it dropped again to 95%.
• During that period, I did not interact with the model or manually start any process.

Expected behavior:

The token/usage counter should remain unchanged if there are no prompts sent, no executed tasks, no user-initiated active sessions, and no manual requests to the model.

Concern:

I am concerned that the Codex token/usage counter is decreasing without any active action from my side. I do not have scheduled tasks, I do not have active automations, and I have not started any model job.

Therefore, I am requesting that you review whether there is any discrepancy between the actual server-side usage recorded for my account and the percentage shown in the interface, or whether the usage meter is incorrectly deducting tokens without any manually initiated request from me.

Please specifically review:

• whether there was actual token usage on my account during that period;
• whether any such usage corresponds to a request manually initiated by me;
• whether there is any difference between the internal usage logs and the percentage shown in the interface;
• whether the Codex counter may be showing an incorrect, outdated, or poorly synchronized value;
• whether the system deducted tokens without any visible user interaction.

Issue details:

• Affected product: Codex.
• Issue type: 5-hour token/usage counter decreasing without active use.
• Initially observed value: 97%.
• Later observed value: 95%.
• User activity during the decrease: no prompts, no chat usage, no task execution, and no active automations.
• Request: please review the internal usage logs for my account during the affected period and confirm whether there was actual usage or whether this is a usage-meter issue.

<img width="1445" height="485" alt="Image" src="https://github.com/user-attachments/assets/dfff2bb9-61db-466b-ade3-33ea9d9dc8da" />

Thank you.

What steps can reproduce the bug?

1. Open Codex.
2. Do not send any prompt or chat message.
3. Do not start any task or model execution.
4. Make sure there are no scheduled tasks or active automations.
5. Observe the 5-hour token/usage counter.
6. Leave Codex open without interacting with the model.
7. Check the counter again after a short period of time.
8. The counter decreases even though no visible user-initiated action was performed.

In my case, the 5-hour usage counter dropped from 97% to 95% without active chat usage, without task execution, and without active automations.

What is the expected behavior?

The 5-hour Codex token/usage counter should remain unchanged when there is no active user interaction.

If I do not send prompts, do not use the chat, do not start any model task, and have no scheduled tasks or active automations, the usage counter should not decrease.

The counter should only decrease when there is actual token-consuming activity initiated by the user or clearly visible in the interface.

Additional information

_No response_

<!-- codex-issue-translator -->

Title

Quota drains rapidly on non-Spark model despite no usage (Pro plan, macOS, Codex v26.609.71450)

Description

I am experiencing unexpected usage quota consumption in Codex after enabling weekly reset.

I also had a weekly reset available and applied it yesterday. It filled my entire weekly and 5-hour limit quota very quickly. Today I used the service very little, but almost 50% of the weekly limit is gone.

In the previous occurrence, my 5-hour window reset at 9:00 PM, and when I returned from dinner around 10:15 PM, the 5 hour usage was already around 13% only, without me using it.

This behavior appears only on the non-Spark model (the other model, 5.3 Codex Spark, does not show this issue). This makes the purchased Pro subscription feel effectively unusable.

Expected Behavior

• Quota should decrement only in proportion to actual usage.

Environment

• OS: macOS
• Subscription: Pro
• Version: 26.609.71450
• Release: Jun 16, 2026

Additional Context

• I can provide screenshots/logs/timestamps if needed.
• This issue is impacting normal usage and subscription value significantly.

Quota drops need visible receipts.

Every decrease in the 5-hour meter should be explainable by a usage event: foreground turn, background task, MCP/server startup, tool retry, compaction, index rebuild, or server-side adjustment. The UI does not need raw prompt text, but it should show timestamp, session/thread id, request id if present, token/cost delta, model, and source class.

If the server ledger has no matching event, the client should show "meter changed, no local usage record" rather than making users infer background activity.

---

_Generated with ax._

Same issue here, down to 80% 5h quota now with 0 prompts sent.

same issue. Codex consume CPU resources and 5h quota drops quickly with limited or no use. this is much worse once I "upgraded" from plus to business.