[macOS 26.908.40834] Generic Chat turn failure is incorrectly shown as “You’ve reached your limit”

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

26.908.40834

ChatGPT for macOS, “Powered by Codex & OWL”, released Sep 11, 2026.

What subscription do you have?

ChatGPT Pro.

What platform is your computer?

macOS 26.5.1, Apple Silicon.

What issue are you seeing?

In a regular Chat conversation in the macOS desktop app, an individual failed turn is intermittently surfaced as:

You’ve reached your limit. Please try again later.

This does not appear to be an actual usage-limit condition.

For the exact same failed turn, opening the same conversation on chatgpt.com shows a generic failure instead:

Something seems to have gone wrong.

Retrying that failed turn on the web can continue to fail, but sending a new plain-text turn in the same conversation succeeds immediately.

In the latest reproduction, the successful subsequent turn synchronized back to the macOS app and was visible there, while the macOS app still displayed the incorrect “You’ve reached your limit” banner.

This initially occurred on a turn involving XLSX attachments, but I later reproduced it on a plain-text-only turn, so it is not specific to newly uploaded files.

The issue looks like incorrect client-side error classification and/or stale conversation error state in the macOS app.

What steps can reproduce the bug?

1. Open an existing regular Chat conversation in the ChatGPT macOS app.
2. Send a normal plain-text message.
3. Intermittently, the turn fails.
4. The macOS app displays:
   You’ve reached your limit. Please try again later.
5. Open the same conversation on chatgpt.com.
6. The same failed turn is shown as:
   Something seems to have gone wrong.
7. Click Retry — the same failed turn may still fail.
8. Send a new simple plain-text message such as test.
9. The new turn completes successfully.
10. Return to the macOS app. The successful turn is synchronized and visible, but the incorrect usage-limit banner may still remain.

What is the expected behavior?

A generic/individual turn execution failure should be shown as a generic retryable error.

The macOS app should only show “You’ve reached your limit” when the backend actually reports a usage/rate-limit condition.

After a subsequent turn succeeds or the conversation is refetched, stale error state should not continue to display an incorrect usage-limit banner.

Additional information

OpenAI Support asked me to submit in-app feedback immediately after reproducing the problem.

In-app feedback session ID:
no-active-thread-01a0a547-2d32-7fd0-8ae0-aeb3896ef77d

Support-provided uploaded thread reference:
019c0d37-d2b6-74c0-918f-0e64af9b6e14

Affected conversation ID:
6a4ec248-77f8-83ed-971c-48beca1c6b9a

Latest reproduced failed turn:
September 15, 2026 at approximately 14:21 Atlantic/Canary (UTC+1) / 13:21Z.

Model/mode: Pro, regular Chat (not Work Mode).

Network: normal home Wi-Fi via Movistar; no VPN, proxy, secure DNS, or corporate network.

Same-account web/macOS comparison

• macOS app: failed turn shown as “You’ve reached your limit. Please try again later.”
• Web: same failed turn shown as “Something seems to have gone wrong.”
• Web Retry: same turn still fails.
• New plain-text message test in the same conversation: succeeds immediately.
• macOS then shows the successful synchronized assistant response, while the incorrect limit banner can remain visible.

Relevant macOS log observations

The desktop log for the affected conversation repeatedly remains in status=error after refetch:

2026-09-15T13:21:18.253Z
chatgpt_conversation_refetch_started
conversationId=6a4ec248-77f8-83ed-971c-48beca1c6b9a
statusBefore=error

2026-09-15T13:21:23.514Z
chatgpt_conversation_refetch_completed
conversationId=6a4ec248-77f8-83ed-971c-48beca1c6b9a
statusAfter=error
statusBefore=error

2026-09-15T13:23:40.268Z
chatgpt_conversation_refetch_started
conversationId=6a4ec248-77f8-83ed-971c-48beca1c6b9a
statusBefore=error

2026-09-15T13:23:45.714Z
chatgpt_conversation_refetch_completed
conversationId=6a4ec248-77f8-83ed-971c-48beca1c6b9a
statusAfter=error
statusBefore=error

2026-09-15T13:38:51.146Z
chatgpt_conversation_refetch_started
conversationId=6a4ec248-77f8-83ed-971c-48beca1c6b9a
reason=missed_messages
statusBefore=error

2026-09-15T13:38:58.925Z
chatgpt_conversation_refetch_completed
conversationId=6a4ec248-77f8-83ed-971c-48beca1c6b9a
reason=missed_messages
statusAfter=error
statusBefore=error

I did not find an obvious HTTP 429 / quota / rate-limit indication in the relevant extracted entries.

An earlier app session around another reproduction also logged:

2026-09-15T12:49:53.372Z
[IpcRouter] Socket error
errorCode=EPIPE
errorMessage="write EPIPE"

I do not know whether the EPIPE event is causal, but it occurred close to another reproduction and may be relevant.

I have screenshots showing:

1. macOS app showing the incorrect “You’ve reached your limit” banner;
2. Web showing the same turn as “Something seems to have gone wrong”;
3. Web Retry still failing for that turn;
4. a subsequent plain-text test turn succeeding in the same conversation;
5. macOS showing that successful synchronized response while still displaying the incorrect limit banner.

I can add those screenshots in a follow-up comment if useful, and can provide the full desktop log files if requested.

Start by reproducing the failure in the macOS app and tracing the conversation refetch events, especially the statusBefore/statusAfter=error logs. Compare the client classification of the failed turn with the web's generic error, then verify that only an actual limit response shows the usage banner and that a successful subsequent turn clears stale error state.