macOS desktop: usage-limit lockout with 24% remaining, followed by restored allowance and access

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

26.903.61454 (build 8378), verified from installed application metadata. Embedded Codex: 0.153.4.

What subscription do you have?

ChatGPT Pro.

What platform is your computer?

macOS / Apple Silicon. Platform output: Darwin 25.6.0 arm64 arm.

What issue are you seeing?

On September 9, 2026, the desktop app blocked use with usage_limit_exceeded after the remaining allowance appeared to fall from approximately 24% to zero. Allowance and access subsequently returned. This report records the temporary lockout and recovery, rather than claiming that 24 percentage points were actually consumed.

A local metadata review verified the following sequence. Times are UTC:

Time	Observation
17:00	Local Codex reports 76% used / 24% remaining in a seven-day window.
17:12:14	A local Codex task fails with usage_limit_exceeded; the adjacent premium quota snapshot has no numerical windows.
17:28:30	A separate desktop Work attempt fails with the same error and missing numerical premium windows.
17:45:43	A new response in the earlier local Codex task succeeds; its codex counter still reports 76% used / 24% remaining.
Subsequent checks	Successful use continues and the live Codex counter reports 77% used / 23% remaining. The user confirms allowance and access have returned.

The rejection message was:

You've hit your usage limit. Visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Sep 13th, 2026 10:35 PM.

The ordinary Codex counter instead resets September 15 at 02:23:46 UTC. The error text did not state its timezone. Missing quota windows are unavailable values, not evidence of zero remaining.

Affected local runtime metadata records GPT-6 Astra / High. There was model/tool processing before the failure; this is not a claim of zero account activity. Local token records are not an authoritative account billing ledger.

What steps can reproduce the bug?

This is an observed incident sequence, not a deterministic reproduction:

1. Use the macOS desktop app while Codex reports approximately 24% allowance remaining.
2. Attempt further local Codex work and a separate desktop Work request.
3. Receive actual usage-limit failures and a retry date inconsistent with the ordinary Codex window.
4. Later, observe a fresh successful local response and remaining allowance again, before the stated retry date.

No synthetic usage-consuming reproduction was performed. No raw session data is attached.

What is the expected behavior?

The allowance display, reset dates, and enforced access should agree. If another limit blocks use, identify that specific limit and its reset instead of showing unexplained exhaustion while another counter remains available. Recovery should leave accurate accounting and usable access.

Additional information

• Verified new feedback receipt: no-active-thread-01a08756-20c9-7573-8d60-216bf0ecb71c
• Earlier feedback reference: no-active-thread-01a08736-3911-7ac0-8c0b-bc62b75039b3
• The new feedback was submitted as a text-only report through the installed Codex App Server feedback/upload API, with logs and attachments disabled.
• OpenAI Status now marks Investigating unexpected usage limit resets resolved. The recovery timing is consistent with that incident, but account inclusion and backend cause are unconfirmed.
• Related Windows evidence: #44234. This report adds the macOS desktop/local-Codex and desktop-Work failure sequence, followed by successful local execution with the earlier allowance still available. Maintainers can determine whether both reports share the same underlying incident.

Please correlate the feedback receipts, reconcile the premium rejections with the codex allowance/reset window, and confirm whether any incorrect deduction or access restriction remains after recovery. A permanent fix or erroneous billing is not established by the available client evidence.

No names, email addresses, account IDs, task/session IDs, local paths, private task contents, raw logs, screenshots, or attachments are included. Feedback IDs are the only supplied personal correlation identifiers.

Start by correlating the supplied feedback/upload API receipts with the local premium quota snapshots, codex counters, and reset dates described in the report. Compare the macOS local Codex and desktop Work failures, including related issue #44234 and the resolved status incident. Done means identifying the enforced limit and reset, and confirming whether accounting or access remains incorrect.