Codex Desktop emits “remote compact task” model-capacity error during automatic compaction

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

26.721.41059

What subscription do you have?

Plus

What platform is your computer?

macOS on Apple Silicon, with Codex Desktop connected to an ARM64 Linux devcontainer. Remote environment: Linux 6.12.76-linuxkit aarch64 Codex runtime 0.145.0

What issue are you seeing?

Automatic context compaction failed twice with:
"Error running remote compact task: Selected model is at capacity. Please try a different model."

The UI first displayed “Context automatically compacted” and then displayed the error. Sending another message caused the same sequence to occur again.

I did not change models. The same task later recovered and continued successfully, demonstrating that changing models was neither necessary nor desirable.

This is especially confusing because the error refers to a “remote compact task” and instructs the user to change models.

A Codex contributor previously stated that compaction had moved locally to the normal-turn endpoint to eliminate remote-compaction errors. Reporting this error again per their comment:
https://github.com/openai/codex/issues/28303#issuecomment-4835137748
"If you see further problems with compaction, please use /feedback and open a new bug report."

What steps can reproduce the bug?

1. Open a long-running task in Codex Desktop connected to a remote devcontainer.
2. Use gpt-5.6-sol with medium reasoning effort.
3. Continue the task until automatic context compaction is triggered.
4. In this occurrence, the preceding turn used approximately 223,423 of the 258,400-token context window (86.5%).
5. Send a normal user prompt and allow Codex to continue working.
6. Observe “Context automatically compacted.”
7. Observe: “Error running remote compact task: Selected model is at capacity. Please try a different model.”
8. Send another normal message without changing models.
9. Observe the compaction message and the same error again.
10. The task subsequently recovers without changing models.

Session ID:
019fa659-bf28-73b1-8d9c-4ad2eb3e14f7

Approximate occurrence:
2026-07-29 06:30–06:38 UTC

Feedback ID:
[PASTE THE ID RETURNED BY /feedback HERE]

What is the expected behavior?

Automatic compaction should complete using the currently selected model.

If a transient capacity failure occurs, Codex should retry transparently or offer a retry action while preserving the selected model. It should not tell the user to change models when the task can recover without doing so.

The task’s tools, pending command state, and context should remain intact.

Additional information

Selected model: gpt-5.6-sol
Reasoning effort: medium
Codex runtime: 0.145.0
App version: 26.721.41059
Subscription: Plus
Remote environment: Linux 6.12.76-linuxkit aarch64
Task originator recorded as: Codex Desktop
Task source recorded as: vscode

At the token event immediately before the failure:

• Context use was approximately 86.5%.
• Account rate-limit use was reported as 0%.
• No rate-limit-reached type was recorded.

The task eventually recovered after a later compaction attempt. No repository changes or pending command output were lost.