codex - 💡(How to fix) Fix Unexpected Usage Draining [2 comments, 2 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
openai/codex#18348Fetched 2026-04-18 05:55:28
View on GitHub
Comments
2
Participants
2
Timeline
8
Reactions
3
Author
Timeline (top)
labeled ×3commented ×2cross-referenced ×2closed ×1

Root Cause

Local rollout logs indicate that the original “Hello World” session completed normally, but additional rollout sessions were created immediately afterward. Those later sessions appears to be unrelated to the original prompt and shows substantially higher token usage, which is consistent with the unexpected usage drain I observed in the VS Code Codex extension. This is especially concerning because the IDE did not clearly show an active follow-up task while usage continued to increase.

Code Example

42377 Apr 17 11:59 rollout-2026-04-17T11-59-20-019d9bf4-1ede-74f0-817b-d2bc4359957a.jsonl
214806 Apr 17 12:00 rollout-2026-04-17T12-00-06-019d9bf4-d364-77f1-9166-3316042337b0.jsonl
167900 Apr 17 12:02 rollout-2026-04-17T12-02-56-019d9bf7-69d0-7da1-b40e-3d636e6977b6.jsonl
RAW_BUFFERClick to expand / collapse

What version of the IDE extension are you using?

26.415.20818

What subscription do you have?

Plus

Which IDE are you using?

VS Code

What platform is your computer?

Linux 6.17.0-20-generic x86_64 x86_64

What issue are you seeing?

In the VS Code Codex extension, usage continues a high draining unexpectedly after a very small task has already finished. I reproduced this with GPT-5.4 Mini on low reasoning by sending only a simple “Hello World” prompt. The response completed normally, but the usage meter kept increasing for several minutes afterward, as if a hidden background task or session was still consuming credits. No visible active thread or running process was shown in the IDE.

The "Hello World!" was sent at 11:59 local time. <img width="613" height="204" alt="Image" src="https://github.com/user-attachments/assets/cb02aa69-115e-4f89-9fec-89c0bf0b1956" />

Local rollout logs indicate that the original “Hello World” session completed normally, but additional rollout sessions were created immediately afterward. Those later sessions appears to be unrelated to the original prompt and shows substantially higher token usage, which is consistent with the unexpected usage drain I observed in the VS Code Codex extension. This is especially concerning because the IDE did not clearly show an active follow-up task while usage continued to increase.

<img width="918" height="74" alt="Image" src="https://github.com/user-attachments/assets/51c2a1b5-a7f9-4dc7-8824-574f33dc881c" />
42377 Apr 17 11:59 rollout-2026-04-17T11-59-20-019d9bf4-1ede-74f0-817b-d2bc4359957a.jsonl
214806 Apr 17 12:00 rollout-2026-04-17T12-00-06-019d9bf4-d364-77f1-9166-3316042337b0.jsonl
167900 Apr 17 12:02 rollout-2026-04-17T12-02-56-019d9bf7-69d0-7da1-b40e-3d636e6977b6.jsonl

What steps can reproduce the bug?

  1. Open VS Code and check the Codex usage meter.
  2. Start a new chat/thread in the Codex extension.
  3. Select GPT-5.4 Mini with low reasoning.
  4. Send a minimal prompt such as “Hello World”.
  5. Wait for the response to complete.
  6. Observe that the usage meter continues increasing for some time even though the task has already ended and no active process is visible.

What is the expected behavior?

A trivial prompt should consume only a very small amount of usage, and usage should stop increasing once the task is finished. If anything is still running in the background, the IDE should clearly indicate it.

Additional information

I recorded a video showing the issue and will attach it to this report. The test was intentionally minimal to isolate the problem. I only reproduced this in the VS Code Codex extension so far, so at the moment it appears to be IDE-related.

Video: https://www.youtube.com/watch?v=ZBR5zBVM5Jw

extent analysis

TL;DR

The issue can likely be mitigated by investigating and potentially terminating hidden background tasks or sessions created by the Codex extension in VS Code.

Guidance

  • Review the rollout logs to understand the pattern of additional sessions being created after the initial prompt, as these seem to be consuming credits unexpectedly.
  • Investigate if there's a way to manually terminate or cancel these background sessions within the Codex extension or through VS Code settings.
  • Consider reaching out to the Codex extension support or developers to report the issue and inquire about any known problems or fixes related to background task management.
  • Verify if the issue persists with different models or settings within the Codex extension to isolate if it's specific to GPT-5.4 Mini with low reasoning.

Example

No specific code snippet is applicable here as the issue seems related to the extension's behavior rather than user-written code.

Notes

The issue appears to be specific to the VS Code Codex extension and might not be present in other IDEs or environments. The provided video and logs are crucial for further investigation.

Recommendation

Apply workaround: Given the information, it seems prudent to look for a workaround such as manually managing or terminating background sessions until a fix is available from the extension developers. This approach allows for continued use of the extension while mitigating the unexpected usage drain.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING