Codex weekly allowance drops by ~1 percentage point per Luna task on ChatGPT Pro
What version of the Codex App are you using (From “About Codex” dialog)?
26.721.81911
What subscription do you have?
ChatGPT Pro ($100/month)
What platform is your computer?
Darwin 27.0.0 arm64 arm
What issue are you seeing?
My Codex weekly allowance is being consumed unusually quickly while using Luna High on a ChatGPT Pro subscription.
Observed usage:
- July 29–30: allowance dropped from 64% to 59% after approximately two hours of work with Luna High.
- July 30: after only 4–6 additional Luna tasks, it dropped from 59% to 55%.
- At approximately 3:02 PM UTC+3, two more Luna tasks reduced it from 55% to 53%.
The latest measured interval therefore consumed approximately one percentage point per task.
This appears comparable to the consumption rate I previously experienced on the $20 Plus plan, despite now paying $100/month for Pro. I would like the metering to be investigated and clarified, including whether retries, tool calls, context size, or background activity are contributing to the deductions.
Related Feedback IDs:
- 019e01f1-1c88-7e60-b9dc-35c5a3016566
- 019e6df8-5a5d-7b43-a999-2d39863137c6
What steps can reproduce the bug?
Feedback ID: 019e01f1-1c88-7e60-b9dc-35c5a3016566
What is the expected behavior?
- Open Codex while subscribed to ChatGPT Pro.
- Record the current weekly usage remaining percentage.
- Select Luna with High reasoning.
- Complete two normal coding tasks.
- Open the usage menu again.
- Observe that the weekly allowance has decreased from 55% to 53%.
Expected result:
Usage should be deducted consistently with the documented model cost and should provide a meaningful allowance advantage over the Plus plan.
Actual result:
Two Luna tasks consumed two percentage points of the weekly allowance.
Additional information
The issue has been reproduced multiple times. My allowance changed from 64% to 59%, then to 55%, and finally to 53%.
Related Feedback IDs:
- 019e6df8-5a5d-7b43-a999-2d39863137c6
- 019e01f1-1c88-7e60-b9dc-35c5a3016566
Screenshots of the usage percentages and feedback submission are attached.
5 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
<img width="274" height="352" alt="Image" src="https://github.com/user-attachments/assets/665ace41-af65-4d9e-8d21-4eb812d8b699" />
<img width="289" height="288" alt="Image" src="https://github.com/user-attachments/assets/f1852e98-472b-421f-ab09-4068afb6caae" />
+1 on this, 64% of weekly usage in one day on $20 Plan, usually it was nowhere so aggressive, same workflow but limit was just dropping like flies.
I'm having the exact same problem. A single Luna (high) agent consumes >1% weekly per hour for me on pro 5x.
The same usage is about 6-8% per hour when using Sol (high).
An hour of deepseek-v4-flash consumes about 30c of API credits for the same workflow, which is a trivial cost, and it's supposed to be ballpark equivalent cost-per-task with Luna now according to ArtificialAnalysis.
Even before the price drop, i have regretted each time I use Luna as it has always used so much more of my usage than the token costs would suggest.
Yet some people report they can run Luna for several hours and it not move their weekly.
I think something is broken.
For the "which tasks are eating it" half of this: Codex already logs per-turn token deltas locally, in the
token_countevents inside~/.codex/sessionsrollouts. That's the same data you're reconstructing by hand from the percentage bar.I built a small MIT tool that reads those rollouts and breaks the drain down by task, project, model, and 5-hour window, priced at API list rates as a common currency: https://github.com/joey-io/gauge. It can't explain the plan's quota mechanics (only OpenAI can clarify the ~1%/task rate), but it makes the pattern visible against your own history instead of a two-day-apart percentage readout. Local only, no network. Codex support shipped today, so if your numbers look off, issues welcome.