[Codex app] Weekly quota fell from ~48% to 7% during an 11-minute Astra Ultra Plan run #44208

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

The most recently confirmed version on this same installation is 26.901.51231 — released September 5, 2026. I did not reopen the About dialog immediately before this specific run.

The affected workflow is in the app's Codex interface.

What subscription do you have?

ChatGPT Pro — $200/month, 20x Codex usage tier.

What platform is your computer?

• macOS Tahoe 26.6.2
• MacBook Pro, 16-inch, Apple M5 Max
• 128 GB memory

Model and workflow

• GPT-6 Astra Ultra
• Codex desktop app
• Plan workflow
• The completed Plan displayed Worked for 11m 5s
• The session displayed 3 completed subagents
• The displayed duration includes time spent waiting for me to answer a clarifying question, so the actual model execution time was shorter than 11 minutes

What issue are you seeing?

A single short Plan run caused my remaining weekly allowance to fall from approximately 48% to 7%.

That is a drop of approximately 41 percentage points of the entire weekly quota during one Plan that displayed only 11m 5s of total elapsed work. This was not a multi-hour implementation run; it was primarily a planning task, and part of the elapsed time was spent waiting for my answer.

This is far beyond merely “high usage” and appears to be a serious usage-accounting anomaly, duplicated metering event, incorrect model/subagent weighting, or entitlement problem. It left only 7% usage remaining, with the UI showing the next weekly reset on September 12 at 1:07 AM.

I do not have access to the server-side per-request token or quota ledger, so please verify the exact before/after values and attribution using the submitted diagnostics.

What steps can reproduce the bug?

1. Use the Codex interface in the macOS app with a ChatGPT Pro 20x subscription.
2. Select GPT-6 Astra Ultra.
3. Note the weekly allowance at approximately 48% remaining.
4. Start a Plan task.
5. Answer the Plan's clarifying question during the run.
6. Let the Plan finish; the UI reports Worked for 11m 5s.
7. Check the weekly allowance again. It now shows only 7% remaining.
8. Submit in-app feedback with diagnostic logs enabled.

What is the expected behavior?

A short Plan run should not deduct approximately 41 percentage points of a Pro weekly allowance. Usage should be metered accurately, should not be duplicated across retries or subagents, and should remain reasonably predictable for the highest paid consumer tier.

If Astra Ultra or parallel subagents intentionally carry a much higher multiplier, the app should provide a clear warning and a per-run usage breakdown before or immediately after the task.

Requested investigation and remediation

Please:

• Inspect the server-side usage ledger linked to the Feedback ID below.
• Identify how much usage was attributed to the main Plan and to each subagent.
• Check for duplicated requests, automatic retries, hidden recovery work, context reprocessing, compaction, cache-accounting errors, or any other repeated metering.
• Verify that the account's 20x Pro entitlement was applied correctly.
• Restore the approximately 41 percentage points of weekly allowance that were abnormally deducted, or the exact amount the backend investigation confirms was erroneous.
• Explain the accounting if the deduction was intentional, and add per-task quota transparency so users can understand and avoid this kind of sudden depletion.

Additional information

Feedback ID: 01a08711-c722-7b93-b166-3813ee781b1b

The in-app feedback submission completed successfully with diagnostic logs enabled.

Related reports from the same account:

• [Codex app] GPT-6 Astra Ultra consumed ~50% of a Pro weekly quota in ~80 minutes across two Goal runs #43967 — approximately 50% of the weekly quota consumed across about 80 minutes of Astra Ultra Goal work.
• [Codex app] GPT-6 Astra Ultra consumes ~30% of a 20x Pro weekly quota during a ~1-hour task #43029 — approximately 30% consumed during a roughly one-hour Astra Ultra task.

This report provides a new run, a new Feedback ID, a different Plan workflow, and a much more severe depletion rate: approximately 41 percentage points during an 11-minute displayed run.

Screenshots document the 7% remaining balance, the 11m 5s duration, the selected GPT-6 Astra Ultra model, the completed subagents, and the successful feedback submission.