Automatic context compaction in Codex Desktop is extremely slow, sometimes taking around a full minute

What version of the Codex App are you using?

26.519.41501 (3044)

What subscription do you have?

Pro

What platform is your computer?

macOS 26.5 on a MacBook Pro 16-inch with Apple M5 Max and 128 GB memory.

What issue are you seeing?

Over the last two days, automatic context compaction in Codex Desktop has become extremely slow.

In the representative session I uploaded, Codex reached the automatic compaction step and paused at Context automatically compacted / Automatically compacting context. The longest observed compaction took around a full minute before the agent continued.

This is clearly abnormal compared with normal task flow. Every long-running task can now spend a significant amount of time blocked on context compaction, which makes Codex feel much slower even when the actual coding/editing work is otherwise progressing.

This report is specifically about automatic context compaction latency. The compaction eventually completes in the representative case, but the delay is long enough to disrupt the workflow.

What steps can reproduce the bug?

1. Open Codex Desktop on macOS.
2. Work in a long-running coding session until automatic context compaction is triggered.
3. Observe that Codex pauses at the context compaction step.
4. In recent sessions, this pause can last close to a full minute before the agent continues.

Representative in-app feedback ID: 019e58c3-fca6-75f0-9512-6876983b86b5

I submitted that feedback with current Codex session logs included.

What is the expected behavior?

Automatic context compaction should not routinely block task progress for around a minute.

Expected behavior:

• Automatic compaction should complete much faster in normal usage.
• If compaction is expected to take a long time, the UI should show clearer progress/status.
• If this is caused by backend latency, timeout, retry, or a recent regression in the compaction pipeline, Codex should make that visible in logs or UI.
• Long-running tasks should not repeatedly lose large amounts of time at the automatic compaction step.

Additional information

The attached screenshot shows a representative session where Codex is at the automatic context compaction stage.

<img width="1672" height="536" alt="Image" src="https://github.com/user-attachments/assets/c3984ccc-45bf-4a31-a752-348e2988bfc8" />