Codex resets all usage limits as OpenAI investigates weekend drain reports

• After saying earlier that Codex's fast quota drain was likely tied to fraud and abuse overflagging, thsottiaux's initial investigation post gave way to a Sunday "warroom" as fresh reports kept coming in, according to thsottiaux's Sunday update.
• OpenAI then issued a hard reset of Codex usage limits for everyone, and thsottiaux's reset announcement said that reset could override as many as three banked manual resets some users had saved.
• Background usage was still murky during the incident: thsottiaux's settings explanation said automations, memory, and suggested prompts can all consume quota, with memory and suggested prompts usually taking about 1% of a weekly limit each.
• The reset rollout itself was uneven enough that dejavucoder's screenshot post showed a user still stuck at 19% weekly quota, until jxnlco's follow-up said the account was actually on Codex for OSS.
• OpenAI's public replies added one more concrete detail about how Codex is wired today: thsottiaux's auto-review reply said auto review uses GPT-5.4 and does not route between models.

thsottiaux's first incident post started with a narrow theory, fraud prevention overflagging. By Sunday, thsottiaux's warroom update had the team combing logs again, while thsottiaux's reset announcement disclosed a hard reset that wiped across normal banked-reset logic. You can also trace the hidden quota consumers in thsottiaux's settings breakdown, and the strangest support thread ended with jxnlco's explanation that one missing reset report came from a Codex for OSS account.

The incident changed shape over two days. thsottiaux's first post said some accounts were seeing usage drain faster than intended and tied the likely cause to abuse and fraud controls that "might be overflagging." Later, thsottiaux's follow-up said the team had a 2X usage promotion multiplier in place and "legitimately" could not find a broad product issue.

By Sunday, that confidence was gone. thsottiaux's Sunday update said the Codex team was in a warroom, combing through logs for anything that could lead to increased usage drains. A few hours later, thsottiaux's public reset post said all users would get a usage reset, with mitigations applied but monitoring still ongoing.

The language on scope also shifted. thsottiaux's earlier reply said OpenAI was not finding users impacted "at large," while thsottiaux's later reset thread still described the investigation as active enough to justify a platform-wide hard reset.

One reason the drain reports were hard to reason about is that Codex can spend quota outside the obvious foreground chat loop. In thsottiaux's settings explanation, thsottiaux listed three background consumers:

• automations
• memory
• suggested prompts

That same reply put a rough number on two of them. Memory and suggested prompts each use around 1% of the weekly limit under normal conditions, and all three can be disabled in settings, according to thsottiaux's settings explanation. The duplicate reply in a second settings explanation suggests support was repeating the same explanation across multiple users.

This does not prove those features caused the weekend drain spike. It does show why users were asking what counted against quota during long-running agent work, especially when usage could move without a visible foreground action.

The most revealing product detail came in the reset itself. thsottiaux's reset announcement said the weekend fix was a "hard reset" and that some users had stacked up to three banked resets they could normally apply on their own schedule.

That implies Codex has at least two separate quota mechanics:

1. the live weekly usage meter
2. a manual reset bank that can accumulate over time

thsottiaux's manual-reset follow-up added that users who had burned a reset only hours earlier would get more manual resets after the investigation concluded. thsottiaux's downgrade reply also clarified that moving from Pro to Plus does not create a fresh pool if the higher-tier quota was already consumed.

The user confusion around this was easy to see. dejavucoder's screenshot post showed a UI with 19% weekly quota remaining and a July 2 reset date, while replies like dejavucoder's follow-up said the user had never even seen a reset bank before.

One support thread ended with a narrower answer than the platform-wide incident. After jxnlco's question about the reset bank and several nudges for visibility, jxnlco's explanation said the user who was missing the new reset was "on the codex for oss!"

That suggests at least some reset logic or entitlement behavior differed between the main Codex product and Codex for OSS during the rollout. The tweet does not explain the boundary, but it does explain why one of the most visible "where is my reset?" reports did not map cleanly onto the broader all-users reset.

While answering quota complaints, OpenAI also disclosed one concrete implementation detail about a separate Codex feature. In thsottiaux's auto-review reply, thsottiaux said auto review uses GPT-5.4 and that Codex does not route those reviews across models.

That reply landed in the same weekend window as the warroom update, which makes the incident a rare moment when public support posts exposed both billing mechanics and model plumbing. The same burst of replies also surfaced a broader app update list in thsottiaux's improvements thread, including smoother handling for very long threads, a hoverable navigation rail, expanded settings search, Slack-preserving Markdown copy, and a dedicated Pets panel.