ChatGPT usage view intermittently disappears; repeated refreshes trigger blocking rate-limit modal

[jyongchul]

Summary

The ChatGPT usage view used to monitor Codex/plan usage intermittently fails to appear in Settings. Reloading until the Usage option becomes available can then trigger a blocking "Too many requests / requests are being sent too quickly" modal, which prevents unrelated ChatGPT work in that browser profile.

This has been reproduced with multiple paid ChatGPT accounts in separate Chrome profiles on Windows.

Environment

• ChatGPT web in Google Chrome
• Windows
• Multiple paid ChatGPT accounts, each in a separate Chrome profile
• Usage routes observed:
  • https://chatgpt.com/#settings/Usage
  • https://chatgpt.com/#usage
  • https://chatgpt.com/codex/cloud/settings/analytics#usage

Steps to reproduce

1. Open ChatGPT Settings and navigate to Usage.
2. Occasionally, the Usage tab/option is missing while the other Settings options render normally.
3. Reload the page until the Usage option appears.
4. Repeat this in more than one Chrome profile, or reload frequently while trying to keep the displayed Codex usage current.
5. A rate-limit dialog appears stating that requests are being sent too quickly and access is temporarily restricted for data protection.

Actual behavior

• The Usage option is intermittently absent.
• The direct Usage hash does not always produce a usable Usage view.
• Repeated full-page reloads can trigger a blocking modal.
• The modal prevents other work in the same ChatGPT page/profile until it is dismissed.
• There is no visible Retry-After time or documented safe refresh interval.

Expected behavior

• The Usage route should render reliably whenever the account supports usage reporting.
• Usage data should update in place, without requiring full-page reloads.
• If polling is rate-limited, the page should expose a retry time and use a non-blocking notice.
• OpenAI should document a supported refresh interval or provide a small authenticated usage endpoint for the current account.
• A once-per-minute refresh should be supported for users who need reasonably current Codex usage information.

Impact

Users monitoring limited Codex/ChatGPT allowances cannot reliably know their current usage. Attempting to keep the value current can temporarily block the rest of the ChatGPT interface.

Workaround tested locally

A browser-side workaround now reloads each usage tab once per minute, staggers the first request across profiles, and applies exponential backoff after the rate-limit dialog. This reduces synchronized bursts, but it does not solve the missing Usage option or provide a supported way to obtain current usage.

Screenshots showing the missing Usage option and the blocking rate-limit dialog are available if maintainers need them.