Repository navigation
Weekly Codex limit dropped from ~45% to 0% with no sessions run (Pro, Desktop app) #42765
Description
What version of the Codex App are you using (From "About Codex" dialog)?
[26.901.31953
What subscription do you have?
ChatGPT Pro
What issue are you seeing?
My weekly Codex usage went from ~45% remaining to 0% remaining in a window where I ran nothing.
Timeline (CEST, Sep 4 2026):
- ~10:00–11:00: Usage & billing showed roughly 45% left on the weekly limit
- ~12:00–13:00: same page shows 0% left, reset Sep 7 2:05 PM
- No Codex sessions, cloud tasks, or background runs between those checks. Desktop app only; no CLI, no IDE extension, no web.
Evidence that nothing consumed it on my side:
- Analytics > Usage history shows a near-zero Codex bar for Sep 4. Heavy days were Aug 31 and Sep 1.
- GPT-5.3-Codex-Spark 5-hour window still at 100% left (resets 6:31 PM), so nothing ran on that model this morning either.
- No Work / Excel / agent usage on this account in the last 7 days. [adjust if not true]
Support case 14377883 has given two AI replies asking for more details and stating limits can't be reset; nothing has been investigated. Opening here so the metering team can see it.
What I'd like: an explanation of what was attributed to my account between 10:00 and 13:00 CEST, and whether this is a metering or attribution bug. I have 2 reset credits and am holding them until this is understood.
Screenshots attached: Usage & billing (0% weekly, Spark 100%), Analytics (flat Sep 4), out-of-usage banner.
Activity
github-actions commentedon Sep 4, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Nebulazer123 commentedon Sep 5, 2026
I’m seeing what appears to be the same issue on ChatGPT Plus, on Codex desktop build 26.901.31953.
In my case this happened twice.
First, before I had done any Codex work on this account that day, I opened Codex and the 5-hour quota was already only approximately 50–60% remaining.
I later legitimately used Codex and exhausted that window.
I then waited for the automatic 5-hour reset at approximately 8:08–8:09 PM CDT and did not start any new Codex/Work task afterward.
The supposedly fresh window then behaved like this:
I have timestamped screenshots for the first two:
The official ChatGPT desktop Usage page and ChatGPT web Usage page showed the same values.
During the unexplained drain I was not submitting prompts, the previous task was already stopped, and I had no cloud Work, scheduled, or Voice/Work task running.
I’ve submitted diagnostic information privately to OpenAI Support as well.
Shteinyr commentedon Sep 6, 2026
Additional report with a similar symptom on a personal Pro 20x account, using Work/Codex on macOS.
On September 5, 2026, at approximately 10:21 UTC+7, my main weekly allowance showed 0% remaining after approximately 1.5 weeks of little or no intentional Work/Codex activity.
Before any plan change or Full Reset:
I later changed to Pro 5x and used one Full Reset to meet a deadline. Access returned, but the earlier Pro 20x depletion remains unexplained.
Help Center AI support confirmed escalation to a specialist, but I have not yet received a human response or case reference.
Could the team investigate whether these reports share a usage-accounting or entitlement issue? I can provide account details and original screenshots privately through Support.
I am not claiming a confirmed common cause. The key question is what historical usage was attributed before the original allowance reached 0%.
frankieyue96 commentedon Sep 6, 2026
Same issue here on ChatGPT Plus.
I checked my limits last night at around 22:08, and my weekly allowance still had roughly 18% remaining.
After that point, I did not use Work, Codex, or Astra at all.
When I checked again this morning, the weekly allowance had dropped to 0%, while my current 5-hour allowance still showed 90% remaining.
So this does not look like normal recent usage exhausting the limit. The weekly quota appears to have been depleted sometime after 22:08 despite no new agentic usage from my side.
I also have screenshots of the quota state before/after.
This looks very similar to the other reports here where the weekly allowance suddenly drops to zero while the short-term allowance remains mostly available.
It would be helpful if OpenAI could check the server-side usage/accounting timeline for affected accounts, especially for delayed posting or quota reconciliation.
Also, my Luna reserve allowance did not change during this period. It remained at the same level while the regular weekly agentic allowance dropped from ~18% to 0%.
This makes the discrepancy even stranger, since the depletion appears isolated to the regular weekly allowance rather than affecting all available usage pools.
jonit-dev commentedon Sep 7, 2026
I got the same #43574
EDIT: I figured the issue. If you don't renew your subscription, instead of a nice UX optimized message telling you to renew, it simply sets you at 0% without explanation.
I had to use another AI to debug.
Please improve this flow. Thanks!
chabelbossa commentedon Sep 9, 2026
I'm seeing what appears to be the same issue on my account.
On Sep 9, 2026, my Codex weekly usage limit suddenly dropped from 50% to 1% remaining, even though I wasn't actively running Codex when the drop happened.
My Analytics data does not seem consistent with that level of quota depletion:
Plan usage for Sep 9 is very low compared with the previous days.
Product activity shows only 12 turns on Sep 9 at the time of the screenshot.
I don't see any unusual usage spike that could explain almost the entire weekly Codex allowance being consumed.
The previous days show much higher activity, yet the weekly limit did not behave like this.
I understand that Analytics can be delayed by up to 6 hours, but the magnitude of the quota drop still seems highly inconsistent with the activity shown.
Screenshots attached for reference.
This looks like it could be a usage metering / quota reconciliation issue, rather than actual Codex consumption.
Is there any way to verify the server-side usage records associated with the weekly limit?
DTadmin-git commentedon Sep 9, 2026
happened to me too
kherona commentedon Sep 9, 2026
I just had this on multiple accounts, 85% to 0 instantly. both accounts 20x pro!
ylz92871-dotcom commentedon Sep 9, 2026
Codex usage quota suddenly dropped while usage activity did not justify it
I am experiencing what appears to be an unexplained Codex usage/quota deduction on ChatGPT Plus.
On September 9, 2026, at approximately 12:00–12:20 PM CDT, my Codex remaining usage suddenly dropped from approximately 60% to 29% within a very short period of time.
I was using GPT-5.6, not GPT-6.0, and I had not performed any unusually large or token-intensive Codex workload that would reasonably explain a drop of approximately 30 percentage points.
My Token Activity page also does not show a corresponding spike in usage during that period.
The issue appeared as follows:
This did not appear to be a gradual depletion caused by normal usage. The quota appeared to change abruptly.
I also checked the usage/activity information available to me, and I could not identify any Codex workload that would account for such a large deduction.
I am therefore concerned that this may be related to a usage metering, quota attribution, or server-side accounting issue.
I would appreciate it if OpenAI could investigate the server-side usage ledger for my account during this time period and determine:
I am not requesting a general explanation of how Codex usage limits work. I am specifically requesting an investigation into the account-level usage that caused the abrupt change from approximately 60% remaining to 29% remaining.
I have screenshots showing both the remaining quota and my Token Activity history and can provide additional diagnostic information if necessary.
ylz92871-dotcom commentedon Sep 9, 2026
mukhamaddin commentedon Sep 9, 2026
Adding another report from September 9, 2026: my Codex allowance showed 59% remaining when I finished desktop work, but later requests were rejected with a usage-limit error. I used a reset to resume work.
Public handle:
mzhakuda684727. Plan: ChatGPT Pro. Platform: macOS desktop. The logs show desktop builds26.901.51231and26.903.61454during the incident.Timestamped evidence
All times below are UTC on September 9, 2026.
used_percent=41, i.e. 59% remaining, with a 10,080-minute weekly window.The afternoon weekly snapshot had
resets_at=1789448129(September 15, 04:55:29 UTC). The evening error instead named September 12 as the retry date. A failed-request event reportedlimit_id=premiumwith both usage windows null, while the earlier and subsequent successful snapshots reportedlimit_id=codex. I do not know whether these refer to different allowances or an accounting inconsistency.Activity checked
I had Codex inspect its local session records and the Hermes logs read-only. The daily totals use the accounting-day boundary of September 8 at 19:00 UTC and exclude this diagnostic conversation; they are not UTC-midnight totals:
Unlike the original report, my account therefore was not completely idle immediately after desktop work ended. I am disclosing the VPS activity rather than attributing the entire drop to zero usage. These token counts do not establish a quota conversion rate, and the local audit cannot see OpenAI's complete account-side ledger.
Analytics screenshots
The attached Plan usage chart shows a small September 9 bar compared with September 7–8. The Product activity screenshot shows 918 turns across the selected seven-day period, not 918 turns on September 9. Neither screenshot captures the original 59% balance or the rejection; those come from the timestamped records above. I understand that analytics can lag, so I am not treating the chart alone as proof of incorrect billing.
There is also a reset-reporting discrepancy: shortly after I used the reset, the account endpoint still reported one available reset credit. I cannot determine from that response whether my reset was redeemed or whether another credit remained.
Could the team inspect the account-side usage attribution between 12:53 and 17:27 UTC, explain the different limit identifiers/reset dates, and verify the reset transaction? If an erroneous block consumed a reset credit, please restore it. I am reporting a possible quota/accounting issue, not a confirmed credential compromise.
Screenshots
frankieyue96 commentedon Sep 10, 2026
After logout all my active device and re-login, the problems gone away and it all seems regular.
ariran5 commentedon Sep 15, 2026
Same issue, today my limits x20 down from 3% to 0% in 20 minutes without active session. And I see how limits go quickly last 3-5 days
mizumico commentedon Sep 27, 2026
For everyone landing here with "I didn't run anything": there are three distinct idle-drain mechanisms currently on record, and the isolation test differs for each:
/statusfrom the CLI. Number stops moving → this is yours./statusreads a few minutes apart — spontaneous upward movement means accounting, not consumption.If it turns out to be (1) or (2) and you can't fully control the background consumers, moving heavy local work to pay-per-token contains the blast radius:
[model_providers]in~/.codex/config.toml(any OpenAI-compatible provider,wire_api = "responses"on CLI ≥ 0.122) bills only what your CLI sessions actually do — background app behavior can't touch it.EzioGuevara commentedon Oct 3, 2026
I’m seeing the same issue on ChatGPT Plus/macOS.
My weekly shared usage reset less than 24 hours ago. I have not used Codex or Work since the reset, and scheduled tasks are disabled.
The quota dropped from 100% → 90% → 89%, while Usage Analytics shows no corresponding post-reset activity. Normal Chat is explicitly excluded from this quota.
This appears to be the same quota-metering/attribution issue.