Repository navigation
GPT-6 Astra weekly usage suddenly burning ~5x faster after paid $80 reset #44894
Description
Summary
I am on ChatGPT Pro and have been using GPT-6 Astra heavily for weeks on the same ongoing build/workflow. My weekly Codex allowance is suddenly being consumed at roughly 5x my normal rate.
I also just purchased an $80 paid weekly reset, so this appears to be affecting a freshly reset paid allowance.
Normal baseline
- Model: GPT-6 Astra
- Same ongoing build / similar workload
- Heavy Astra usage for weeks
- Historically, 20% of weekly usage lasts 10+ hours for this workload
Current behavior
- Purchased an $80 weekly reset
- After the reset, approximately 20% of weekly usage was consumed in about 2 hours
- This is roughly a 5x increase in quota consumption compared with my established baseline
- There has been no comparable 5x increase in workload or change to the build that would explain the difference
Why this appears abnormal
This is not a comparison between Astra and another model. I am comparing Astra to my own prior Astra usage over multiple weeks, using the same build and similar working pattern.
There are also other recent reports involving unexpected Codex usage/reset behavior and unusually fast Astra quota consumption, including reports around paid/reset weekly limits.
Expected behavior
A freshly purchased weekly reset should provide the normal weekly allowance, and the same Astra workload should consume quota at approximately the same rate as it has over prior weeks.
Actual behavior
The newly reset weekly allowance is currently draining approximately 5x faster than my established Astra baseline.
Please investigate the server-side usage accounting for GPT-6 Astra and paid weekly resets, including whether recent rate-limit/metering changes or reset-accounting issues are causing excess weekly allowance consumption.
Activity
github-actions commentedon Sep 11, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
vkeezou commentedon Sep 13, 2026
Same class of issue here, adding a concrete before/after dataset.
Plan: ChatGPT Pro (Pro 5x, $100/month), Windows desktop app, local sessions only.
Timeline: after the Sep 1 global reset (the eight token-drain bug fixes) my usage was reset to 100%. Over the following three days (Sep 1–4) I only ran local Codex sessions and the dashboard fell to ~3%.
Verifiable local data for that window (from the local Codex session DB, not the server meter):
Feedback ID: no-active-thread-01a09557-ad2c-7290-88d2-3b7d402f9173
(Account-level ticket also filed via help.openai.com; private account details omitted here.)
Update (2026-09-12) - correcting a plan detail in my earlier comment: my Pro subscription ran
Jul 28 - Aug 28, 2026 and was not renewed afterwards, so the account has been on the Free plan
since Aug 28. The paid-month data is: 119 sessions, 1,122,966,107 tokens, estimated
70,000-123,000 credits (about 90% on gpt-5.6-sol; about 49% of the estimated cost from
subagent-spawned sessions; peak day Aug 4 about 29,800 credits). I waited for the reset date
shown in the app (Aug 20) and no allowance returned, so I stopped using Codex and did not renew.
OpenAI granted my account a banked reset on Aug 22 07:57 (GMT+8); I only saw it on Sep 3 and
redeemed it at 16:28. The Sep 1 full reset for paid subscribers did not reach me because my
subscription had lapsed on Aug 28 - a lapse caused by the incident. The "100% to 3% in days"
figure I mentioned earlier relates to the Free plan's small allowance afterwards, not to faulty
metering, and I am not presenting it as such.
pedrometer12 commentedon Sep 16, 2026
I have experienced what appears to be a closely related problem, with purchased credits being consumed at an extremely rapid rate that I cannot reconcile with either the work performed or the usage analytics. I have documented my case here: #45932
I have also found similar reports at #36622 and #45688. In my case I have raised the disputed usage with OpenAI Support but have not yet received an explanation or had the credits restored.
rodraco commentedon Oct 14, 2026
Same here, on the 100 max plan - Astra on high only for planning, sol-6.1 code + test - cannot get even half work done compared to before; openai did not respond to any question, same topic on the forum was closed without an explanation; i think too many signed up for this, time to free up some capacity moving back.