OpenAI resets Codex plan limits after rate-limit spike and fraud bans

• OpenAI reset Codex usage limits across all plans after internal dashboards showed more users than expected hitting caps, and the company said it still does not fully understand the cause Tibo Sottiaux on the reset.
• User screenshots show the reset was visible in the product, with 5-hour and weekly balances returning to 100% on Pro accounts Pro dashboard after reset.
• The bigger issue is not the free extra quota. It is that Codex has a layered limit system, across 5-hour windows, weekly code review caps, model-specific allowances, and optional credits, which makes metering bugs hard to reason about from the UI alone reset screenshot, OpenAI Codex pricing.
• This was not an isolated blip. GitHub issues from March describe weekly and 5-hour quotas draining together or faster than expected, and one commentary tweet claimed OpenAI had already reset Codex limits six times in March claim of repeated resets, rate-limit issue #15039, issue #14628.
• OpenAI also said it banned a batch of fraudulent accounts and regained compute, which suggests at least part of the pressure was supply-side, not just a UI or accounting mistake fraud-ban explanation.

The useful reveals are pretty direct: OpenAI publicly admitted the quota behavior is still not fully explained Tibo Sottiaux on the reset, users immediately saw their balances refill in the dashboard Pro dashboard after reset, and the official pricing page shows why this keeps getting messy, because Codex mixes shared 5-hour windows, weekly review caps, model-specific limits, and pay-to-continue credits in the same product OpenAI Codex pricing. The weird part is the fraud note. OpenAI did not just blame demand, it said abuse bans freed up compute, which hints that quota pressure and capacity management are still tightly coupled in Codex fraud-ban explanation.

The core fact here is simple. OpenAI reset Codex usage limits for all plans after seeing an increased rate of users hitting rate limits, and it did so before it had a clean root cause.

That matters more than the free refill. For engineers using Codex in production workflows, this is a direct signal that the quota system is still being tuned live. If your team treats the dashboard as a stable capacity contract, this tweet says not to.

The user evidence lines up with the announcement. Screenshots from Pro users show both session and weekly quotas back at 100%, which rules out this being a narrow support-side adjustment or a plan-specific fix Pro dashboard after reset.

Codex does not have a single meter. The official pricing page breaks usage into local messages, cloud tasks, and GitHub code reviews, with different limits by model and plan.

A few details matter:

• GPT-5.3-Codex on ChatGPT Pro is listed at 300 to 1500 local messages per 5 hours, 50 to 400 cloud tasks per 5 hours, and 100 to 250 code reviews per week.
• GPT-5.4 and GPT-5.4-mini use a different structure, with local-message limits but no cloud-task or code-review bucket on the pricing table.
• OpenAI says local messages and cloud tasks share a five-hour window, and additional weekly limits may still apply.
• After you hit included limits, Plus and Pro users can buy credits to keep going.

That structure helps explain why users get confused. The dashboard can show a couple of friendly progress bars, but the underlying system changes by model, task type, and execution mode. The help article also warns that message counts vary with task size, codebase size, and how much context Codex has to hold, so two teams doing the same number of prompts can burn very different amounts of allowance.

The April 1 reset landed on top of a backlog of rate-limit complaints, not a calm baseline. In issue #15039, a user reported that 5-hour and weekly quotas were being consumed at almost the same rate, even though they expected those limits to behave independently. In issue #14628, another user said support acknowledged rate-limit consumption that did not match actual usage.

That does not prove every complaint had the same cause. It does show a pattern: users have been struggling to map visible usage to actual work for weeks.

One commentary tweet, relaying another post, claimed OpenAI had already reset Codex limits six times during March claim of repeated resets. Even if that count is approximate, the broader point looks right. Resetting usage has become a recognizable part of the product's operating rhythm, enough that users now joke about it as a recurring event community reaction, employee reply.

The most revealing line in the reset tweet was not about generosity. It was the admission that OpenAI found a pocket of fraudulent accounts, banned them, and regained some compute.

That tells you the quota story is also a capacity story. Limits are not only there to meter legitimate demand. They are also acting as a buffer against abuse and a control surface for scarce compute. When abuse drops, OpenAI can hand some of that headroom back to paying users. When usage spikes or metering drifts, the same users slam into caps faster.

There is also some product history behind this. When OpenAI launched the Codex app, it said it was doubling rate limits on Plus, Pro, Business, Enterprise, and Edu plans for a limited time, across the app, CLI, IDE, and cloud. A product with growing distribution, temporary limit boosts, mixed local and cloud execution, and active abuse pressure is exactly the kind of system where quota math gets harder before it gets cleaner.

For now, the practical read is narrow: watch the dashboard, check /status during active CLI sessions, and assume visible quota behavior can still change faster than the docs do.