FULL STORY

OpenAI's Quota Glitch: From User Complaints to Official Probe

Starting late August, Codex and ChatGPT users reported abnormal quota depletion with no explanation from OpenAI. After a surge of Codex resets on September 10, OpenAI confirmed an incident and launched an investigation.

2026-08-25 ~ 2026-09-11 · 5 episodes · 32 posts

Episode 1 · Users Suspect OpenAI Silently Cut Codex Plus Limits (2026-08-25, 4 posts)

Multiple users report Codex Plus weekly limits draining abnormally fast, some exhausting quota in two days despite unchanged workflows, and suspect OpenAI silently reset or cut quotas without any official announcement—an employee only admitted they "forgot to mention it."

Episode 2 · OpenAI Codex Usage Limits Force Open Source Dev to Pause Work (2026-09-06, 2 posts)

Developer cnakazawa reported that after his OpenAI Codex OSS grant expired, Premium Business usage limits left him with only 60% weekly quota and no way to buy resets, prompting him to halt Codex-related open source work for a week.

Episode 3 · ChatGPT Users Report Quota Draining Abnormally Fast (2026-09-09, 4 posts)

Multiple ChatGPT users report quotas draining abnormally fast—some hitting limits after just a few prompts—along with a quietly delayed weekly reset, with no adequate explanation from OpenAI.

Episode 4 · Codex Users See Usage Meters Wiped from 92% to Zero (2026-09-10, 2 posts)

Codex users report their usage meters dropping abruptly from 92% to 0%, suspecting a billing glitch; OpenAI staff attributed it to a display issue amid earlier similar anomalies.

Episode 5 · OpenAI confirms bug behind abnormal Codex usage limit resets (2026-09-10, 20 posts)

On September 10, numerous Codex users reported sudden abnormal drops in their usage quotas. OpenAI posted an incident notice on its status page confirming "unexpected usage limit resets" for some Codex users and opened an investigation; multiple staff members confirmed it was a bug, with engineer Thibault Sottiaux calling it a known issue being fixed. The event was attributed to a metering defect rather than actual consumption or policy tightening, and some scenarios (such as quota drain during sleep) were already patched.

Confirmed

  • OpenAI's status page posted an incident notice around 17:29 UTC Wednesday about abnormal usage limit resets for some Codex users, with investigation underway and no specific affected component named.
  • Staff members pvncher and jdjohnson separately acknowledged anomalies in Codex usage limits and said the team was working on it, both asking users not to spam unrelated posts with support requests.
  • Per carlosdponx's follow-up, the Codex team first said it was verifying and escalating, then confirmed it was a bug being fixed.
  • Per shaunralston's relay, engineer @thsottiaux responded to AIandDesign's 20x plan complaint calling it a known issue under resolution.
  • Developer danielmac8 found quota draining while his laptop was asleep (85% before, 40% after lunch); after he reported it to @OpenAIDevs, it was quickly fixed.
  • User reports include: koltregaskes dropping from 30% to 0; DimitrisPapail from 75% to 1% in minutes on weeks-old routine tasks with Astra; AIandDesign from 55% to 8% in 30 minutes on MEDIUM with Astra on a 20x plan; gpt2chatbot's Codex Pro from 70% to 17% in two hours (with screenshots, noting five Astra Max Fast agents over a weekend consumed less); jamesbrooksco exhausting a fresh reset within 30 minutes without Astra and publicly questioning Thibault Sottiaux; carlosdponx from 100% to 36% within 10 minutes with reset shown three days out, plus a relayed case dropping from 92% to zero; yuntiandeng down to 20% just 9 minutes after a manual reset.
  • Self-recovering displays: dotey saw a drop from 70% to 7%, found actual usage normal in settings, and watched it restore to 75%; pwlot was zeroed from 39%, paid for a reset, then saw it return to 35% within minutes—nearly the pre-zero level—prompting him to question whether it was a bug or a scheme.
  • In BLUECOW009's roundup, xikhar fell from 40%+ to 0% with the reset date silently moved from September 15 to September 12; BLUECOW009's own quota fell from 90% to 10% overnight. RileyRalmuto noted his timeline was flooded with such complaints while he had used only 9%, lamenting the next day that the "Codex reset doom" came at a bad time.

Unconfirmed

  • OpenAI has not explained the technical root cause, the scope of affected users, or whether users like pwlot who paid for resets due to false displays will be compensated; whether perceived "quota shrinkage" stems from this bug or longer-term quota adjustments also remains unanswered.

Why it matters

  • The incident hit paying and high-tier users (20x plans, Codex Pro, ChatGPT Plus) in a short burst, pointing to a systemic metering defect rather than isolated cases—though staff confirmed and fixes are in progress.
  • dotey's and pwlot's experiences show some drops were display-layer bugs that self-recovered, yet users already paid real money based on false readings, affecting actual bills.
  • Users lack usage charts and quota alerts, making consumption hard to audit; gpt2chatbot publicly called on OpenAI to add visualization, and replies under jdjohnson's post noted that poor official support channels drove users to Twitter, amplifying product and service shortcomings.