Usage limits appear to be reduced without the promised community engagement and transparency #44698

Activity

1. kendonB

   added

   bugSomething isn't workingSomething isn't working

   on Sep 11, 2026
2. github-actions

   added

   CLIIssues related to the Codex CLIIssues related to the Codex CLI

   rate-limitsIssues related to rate limits, quotas, and token usage reportingIssues related to rate limits, quotas, and token usage reporting

   on Sep 11, 2026
3. github-actions commented on Sep 11, 2026

   

   github-actionsBot

   on Sep 11, 2026 – with GitHub Actions

   Contributor

   More actions

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

   • Pro 20x: Astra shows ~36% lower API-equivalent capacity per weekly quota than Sol (same-account measurements) #43731
   • Severe Codex Weekly Quota Depletion / Possible Usage Accounting Bug #43811
   • After deploying ASTRA #44653

   Powered by Codex Action
4. kendonB commented on Sep 11, 2026

   

   kendonB

   on Sep 11, 2026

   Author

   More actions

   Original source for the 14K

   
5. austindtranz-cloud commented on Sep 11, 2026

   

   austindtranz-cloud

   on Sep 11, 2026 – with ChatGPT Codex Connector

   More actions

   I’m a ChatGPT Pro subscriber using Codex on Windows. As of September 11, GPT-6 Astra is consuming my allowance much faster than I experienced with GPT-5.5, even without Ultra reasoning. The current consumption rate makes sustained everyday work difficult to plan.

   I understand that Astra has higher published usage rates than Sol. However, I would like to know whether the depletion I’m experiencing is entirely expected or whether an accounting issue is still affecting my account.

   The September 7 automatic reset and the resolution of the September 9 usage-limit incident have not resolved my ongoing concern about the burn rate.

   Could OpenAI please clarify:

   1. Is there a known ongoing issue with usage accounting, cached-input charging, or quota reconciliation?
   2. Have the included allowances or model-specific usage weights changed since GPT-5.5 and the Astra launch?
   3. Can Support review my account’s server-side usage accounting and explain what is consuming the allowance?
   4. If this is a bug, what is the expected timeline for a fix, and will incorrectly deducted usage be restored?

   I also want to raise a product concern: I prefer one reliable default model for everyday work. Constantly switching between Astra, Sol, Terra, and Luna to manage an opaque shared allowance adds friction. Is Astra intended to be usable as a daily default on Pro, or only for occasional demanding tasks?

   Please provide a concrete explanation and practical guidance beyond switching to a smaller model or purchasing more credits. I can provide my account ID privately to Support, but I do not want to publish personal information, logs, or private task content.

   👀React with 👀1Reacted by Kendon Bell
6. austindtranz-cloud commented on Sep 15, 2026

   

   austindtranz-cloud

   on Sep 15, 2026 – with ChatGPT Codex Connector

   More actions

   Follow-up with additional sanitized context from September 15, 2026:

   This issue has now recurred across two separate paid accounts. Both accounts received a reset around September 12. The main account exhausted its renewed weekly allowance again by September 15, in less than three days. The second account had previously exhausted its renewed allowance in approximately two days. I am omitting account emails, order references, and private identifiers from this public comment.

   Before GPT-6 Astra, including GPT-5.5 and GPT-5.6, my weekly allowance normally lasted the full week for my usual client-development workflow. I rarely needed to change models. During the recent period I had to switch repeatedly between Astra, Sol, Terra, GPT-5.5 and Luna simply to keep work moving. This is not my historical usage pattern.

   The five previously reported Astra High turns on September 11 showed:

   • 23 responses, 2,452,273 input tokens, 2,371,456 cached input, 11,003 output, over about 9 minutes.
   • 48 responses, 8,910,089 input, 8,681,344 cached input, 24,934 output, over about 21 minutes.
   • 16 responses, 4,071,377 input, 4,044,800 cached input, 6,758 output, over about 6 minutes.
   • 21 responses, 7,433,502 input, 7,410,048 cached input, 13,587 output, over about 13 minutes.
   • 10 responses, 4,877,597 input, 4,822,656 cached input, 20,370 output, over about 14 minutes.

   Cached input represented roughly 96.7%–99.7% of input in those examples. These are local model counters, not claims about exact account-level percentage debits. Other turns and subagents overlapped, so only OpenAI can reconcile them with the server-side allowance.

   During the affected period, experimental context management was enabled (features.context_management.experimental_mode = true) with a 1,000,000-token context window. It has since been disabled. I do not know whether history retrieval, notes, context transitions, or reprocessing contributed to allowance consumption. Please clarify whether this experiment can affect subscription usage accounting or cached-input treatment.

   Fast mode was generally off (service_tier = default). When tested, I did not perceive a meaningful improvement in reasoning wait time, code generation, or total completion time despite higher consumption.

   The weekly-limit structure itself is not my complaint. Removing the five-hour restriction improved the workflow. The concern is the practical weekly capacity and the lack of transparent accounting when Astra, model switching, Ultra orchestration, subagents, retries, compaction and context management are involved.

   Questions for OpenAI:

   1. Can the repeated exhaustion across two accounts be correlated with a server-side accounting or rollout issue?
   2. What exact usage does Pro 5x versus Pro 20x provide for weekly Astra work, rather than only the five-hour message estimates?
   3. How are Astra Ultra parent runs and child agents attributed to the shared allowance?
   4. Can experimental context management increase usage or bypass cached-input treatment?
   5. If the accounting is correct, is OpenAI planning changes that make Astra sustainable for ordinary daily client work?

   This is a sanitized public supplement to the private Support case. No private code, raw logs, account identifiers, or task transcripts are included.
7. kendonB commented on Sep 29, 2026

   

   kendonB

   on Sep 29, 2026

   Author

   More actions

   Thanks sir @tibo-openai https://x.com/thsottiaux/status/2104823812042940713
8. kendonB

   closed this as completedon Sep 29, 2026