Codex 5-Hour Quota Consumed Disproportionately During a Single Long-Running Workflow

Open 💬 2 comments Opened Jun 18, 2026 by excessive-eng

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

Version 26.611.62324

What subscription do you have?

ChatGPT PRO 5x

What platform is your computer?

Darwin 25.6.0 arm64 arm

What issue are you seeing?

I am observing rapid depletion of the Codex 5-hour quota during a single continuous workflow.

On June 18 I started a single Codex task involving document digitization, comparison, OCR verification, Excel processing and report generation.

The task remained within one conversation and one workflow thread. I did not continuously submit large numbers of new prompts.

However, the 5-hour quota dropped unusually fast:

  • approximately 97% remaining
  • approximately 90%
  • approximately 83%
  • approximately 77%
  • approximately 74%
  • approximately 48%
  • approximately 46%
  • eventually: “You’re out of Codex messages”

The quota exhaustion occurred while working within the same task and appears disproportionate to the number of user messages sent.

Why this looks suspicious:

  • the work was a single continuous workflow
  • no parallel Codex sessions were intentionally running
  • no Codex CLI usage
  • no GitHub integration
  • quota consumption appears tied to internal task execution rather than visible user prompts

Additional evidence:

I collected local system monitoring logs during the same period.

The logs show:

  • long-running background activity
  • high system load averages for extended periods
  • no obvious increase in user interaction frequency
  • no indication that dozens of separate user requests were submitted

Expected behavior:

  • quota consumption should be predictable and proportional to actual user interaction
  • Analytics should expose which internal operations consume quota
  • users should be able to distinguish between:
  • user prompts
  • internal agent loops
  • retries
  • background processing
  • tool execution
  • autonomous task continuation

Requested investigation:

Please review:

  • quota accounting for this session
  • number of internal model invocations generated from the workflow
  • whether autonomous agent steps consume quota identically to user prompts
  • whether long-running internal workflows can unexpectedly exhaust the 5-hour quota
  • whether the displayed remaining quota accurately reflects actual user message consumption

I can provide screenshots showing the quota progression during the session and local monitoring logs collected throughout execution.

What steps can reproduce the bug?

  1. Open Codex in the ChatGPT desktop application.
  2. Start a normal working session involving document analysis, file processing, report generation, and Excel-related tasks.
  3. Work within a single conversation and a single workflow thread.
  4. Periodically monitor:
  • remaining 5-hour quota;
  • remaining 7-day quota;
  • session ID;
  • context size.
  1. Continue working on the same task.

Observed behavior:

09:14
5-hour quota remaining: 97%
7-day quota remaining: 100%

09:21
5-hour quota remaining: 90%
7-day quota remaining: 98%

09:28
5-hour quota remaining: 83%
7-day quota remaining: 97%

09:40
5-hour quota remaining: 74%
7-day quota remaining: 96%

10:05
5-hour quota remaining: 49%
7-day quota remaining: 92%

10:17
5-hour quota remaining: 46%
7-day quota remaining: 92%

11:14
Codex became unavailable with the message:
“You’re out of Codex messages”

The issue occurs during a single continuous workflow and appears disproportionate to the number of user prompts submitted.

Additional observation:

The same behavior has occurred multiple times. In previous cases, quota continued decreasing while context size remained unchanged and while little or no visible user activity was taking place.

I cannot provide a deterministic reproduction sequence because the issue appears to be related to quota accounting or internal task execution rather than a specific prompt. However, it has now occurred repeatedly during normal Codex usage.

What is the expected behavior?

_No response_

Additional information

_No response_

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗