Repository navigation
Erratic Usage Limits & Resets Changing Every Other Day #36307
Description
What issue are you seeing?
I have a codex subscription and it used to be reliable but for the past 2 weeks I have been unable to use it.
The weekly limit gets used up in a matter of hours when I have it, and over the past 2 weeks the limit reset has jumped every other day basically from:
19th July > 23rd > 25th > 28th > 1st Aug > 5th Aug rendering the service completely unusable, unpredictable and erratic leading me to cancel my subscription and pay for other services. I have been a paying customer since 2023 and have never seen the service be so unreliable before.
What steps can reproduce the bug?
To use the app as I usually do
What is the expected behavior?
To not use a weekly limit in a few hours, and to not have my weekly limit reset jump every few days being erratic and unreliable
Additional information
Such a shame as I used to really rely on your service but now cant.
Activity
github-actions commentedon Jul 31, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
phyrexia commentedon Jul 31, 2026
Same here — second independent account, with client-side telemetry.
I hit this over Jul 28–31 on a paid plan. What may be useful: the Codex CLI writes a
rate_limitsblock on every turn (used_percent,window_minutes,resets_at,plan_type), so I could reconstruct the series rather than rely on screenshots. Values below are read verbatim from that field; the counts are mine, over the stated window.1) Your reset dates match mine. Over 48h my account reported NINE distinct weekly windows (after collapsing sub-second jitter):
Four are on your list (Jul 25, Jul 28, Aug 1, Aug 5). Two accounts, different machines, same anchors.
2) A brand-new account burned its entire weekly quota on ONE task. After exhausting my resets I created a fresh account. Per Codex's own
/usage daily, its whole lifetime is: 374M tokens,Streak 1d, longest task 1h40m — lifetime equals peak, because that single task is all the account ever did. One task, and the week was gone.3) I redeemed all three of my weekly reset credits, and none of them is auditable. They do not appear in
/usage daily, and they do not appear in the local telemetry either: there is no field that ties a quota reading to a specific window or credit (limit_idis the literal stringcodexfor essentially every reading, andlimit_nameis always null). A user cannot verify what a reset actually restored.4) The window looks recomputed rather than anchored — which matches #35952. Of the anchors I can test against my own activity, each falls exactly 7 days after a moment I was actively using Codex, to the minute:
5) Expired windows keep filling instead of rolling over. Three of the nine had a
resets_atalready in the past (Jul 25, 27, 28) and were still receiving usage on Jul 29–30, climbing to 100%.6)
window_minuteswas 10080 (7 days) in 100% of readings in that period.7) Over Jul 28–30 my client logged 282
usage_limit_exceedederrors, against 2 on each of the two preceding days.Happy to share the extraction method — it only reads fields the CLI already writes locally.
douglascvas commentedon Aug 1, 2026
Same with me. With GPT 5.4 and 5.5 I was able to code for a whole week and would be left with 30 to 50% of my quota. Now I am coding less and my Pro 20x quota is gone in 1 day.
johentsch commentedon Aug 11, 2026
Same here, completely erratic. It was going to reset on the 11th of August. When I had used up a bunch, the reset shifted to the 15th. Then I was reaching the limit mid-task so I had to use the last reset available (the number of available resets also seems to be unreliable), after which the limit shifted to one week later and I could finish my tasks. Now I come back and -- TADAA -- limit has been reset AGAIN. How am I to take this service seriously?