Limits erroneously drawn to 0% in less than a few minutes when not using codex

Open 💬 4 comments Opened Jul 10, 2026 by avinashresearch1

💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

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

ChatGPT Powered by Codex & OWL Version 26.707.31428 Released Jul 10, 2026

What subscription do you have?

Pro

What platform is your computer?

macos

What issue are you seeing?

I barely used Sol-light. And it drew the usage down to 45%. Then I completely stopped using it at all. The usage moved from 45% to 0% when I was not even using the app at all. There was nothing running in the background as far as I could tell.

Case Number: 11276914 in support
Feedback ID: 019f4d90-627f-7873-a01b-f382603baa99

│ │ - Codex │ - gpt-5.4-mini │ 1,805,627 │ 110,189 │ 0 │ 27,983,232 │ 29,899,048 │ $24.27 │
│ │ │ - gpt-5.6-luna │ │ │ │ │ │ │
│ │ │ - gpt-5.6-sol │ │ │ │ │ │ │

What steps can reproduce the bug?

Feedback ID: 019f4d90-627f-7873-a01b-f382603baa99

What is the expected behavior?

I was hoping to use codex at least to the level of 5.5xhigh where I could have many turns. Major regression.

Additional information

_No response_

View original on GitHub ↗

4 Comments

github-actions[bot] contributor · 1 month ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

#30943
#31796
#31402
#31784
#32250

Powered by Codex Action

avinashresearch1 · 1 month ago

So, investigating the logs. I made the following changes which seems to make things significantly better:

The sub-agent use does not properly cache tokens. I fully disabled all use of sub-agents in my system prompt and config file. Investigating logs told me that the initial large chunk of tokens was resent 4 times to the different agents and this was the cause of the leak.
Memories take up an absurd amount of tokens. I disabled memory and chronicle. I believe this was the primary cause of "idle" usage.
I changed my system prompt to encourage the agent to be less thorough: I previously had specified that it must read and understand the codebase and write research files first. That previous instruction completely blew my usage.

These 3 fixes seem to give me reasonable usage.

gokdeniz-cakir · 1 month ago

I have had the exact same issue today, the high usage of the new models aside, my 5h limit drained from 47 to 0 without any turns inbetween.

JCBrown · 1 month ago

I saw this issue this morning, no turns overnight [*1] however 68% 7 day limit is showing 0%. I've previously wondered why usage can drain quickly and attributed it to subagents having to load significant context.

I've disabled Memory as-per previous comment and will see how I do with a usage reset...

Of note, the Status command no longer shows any 5h limit... [*2]
<img width="785" height="206" alt="Image" src="https://github.com/user-attachments/assets/19daa711-e5e1-4e50-aa71-3a722aa290dc" />

The behaviour I'm seeing seems very similar to that reported by https://github.com/openai/codex/issues/31668#issue-4841872779

[*1] as verified via session logs

[*2] Update

OpenAI has removed the 5-hour limit; this is not a bug. Closing this now.

https://github.com/openai/codex/issues/32707#issuecomment-4955180935