Erratic Usage Limits & Resets Changing Every Other Day

[Sammii-HK]

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.

[github-actions]

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

• Weekly limits reset extending to infinity when we are getting close to it #35952

Powered by Codex Action

[phyrexia]

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_limits block 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):

Jul 25, Jul 27, Jul 28, Jul 31, Aug 1, Aug 4, Aug 5 (x2), Aug 6

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_id is the literal string codex for essentially every reading, and limit_name is 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:

reset Aug 5 05:30Z  =  my activity Jul 29 05:30Z + 7d
reset Aug 5 19:07Z  =  my activity Jul 29 19:07Z + 7d
reset Aug 6 13:10Z  =  my activity Jul 30 13:10Z + 7d

5) Expired windows keep filling instead of rolling over. Three of the nine had a resets_at already in the past (Jul 25, 27, 28) and were still receiving usage on Jul 29–30, climbing to 100%.

6) window_minutes was 10080 (7 days) in 100% of readings in that period.

7) Over Jul 28–30 my client logged 282 usage_limit_exceeded errors, 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]

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.