OpenAI resets Codex usage limits across all plans after a rate-limit spike

• OpenAI reset Codex usage limits across all plans on April 1 after internal dashboards showed more users hitting caps, and Codex product lead Tibo Sottiaux said the team did not yet fully understand the trigger Tibo Sottiaux on the reset.
• Sottiaux also said OpenAI had banned a pocket of fraudulent accounts and recovered some compute, tying the reset to both demand pressure and abuse cleanup Tibo Sottiaux on fraud accounts.
• User screenshots show the reset restored both short-window and weekly allowances back to 100%, not just a single session bucket User screenshot of restored limits.
• The reset landed after a month of repeated limit changes, with community members counting multiple March resets and GitHub users complaining that weekly reset dates kept moving Community count of March resets.

You can read OpenAI's plan guide for Codex, the pricing page that spells out shared five-hour windows and weekly caps, and a fresh GitHub issue from a power user whose weekly reset jumped from April 3 to April 8. The funny bit is that one user assumed the reset was an April Fools joke, then an OpenAI employee replied, "We don't joke about rate limits" April Fools confusion OpenAI employee reply.

The key fact is unusually direct: OpenAI reset limits first and explained later. Sottiaux wrote that Codex dashboards were showing an increased rate of users hitting limits, and that the team still did not fully understand why.

That made the move less like a scheduled quota refresh and more like an operational stopgap. Within minutes, users were posting screenshots showing their Codex balances back at 100% on Pro accounts Pro dashboard back at 100%.

Sottiaux's post bundled two explanations together: unexplained pressure on rate limits, and a newly discovered cluster of fraudulent accounts that OpenAI banned to regain compute Tibo Sottiaux on fraud accounts. The combination suggests the extra headroom came partly from cleanup, not from a neatly planned expansion.

OpenAI's own documentation already frames Codex as a constrained service. The pricing page says local messages and cloud tasks share a five-hour window, weekly limits can also apply, and limits vary by model and plan. The newer rate card also points to a flexible-pricing credit system across Plus, Pro, Business, and Enterprise or Edu plans.

By April 1, users were already treating another reset as business as usual. Replies called Sottiaux the "saint of codex rate limits" and joked that "reset first talk later" had become the standard move User reaction to repeated resets Reset first talk later.

The stronger evidence is outside X. In the OpenAI developer community, users reported that weekly reset days were drifting and that March 27 and April 1 both came with fresh credits. On GitHub, one contributor wrote that a weekly limit changed from 60% remaining until April 3 to 100% remaining until April 8, which scrambled how they rationed heavy workflows like agent swarms and fast-mode sessions in the second half of the week, GitHub issue.

One useful detail in the screenshots is structural. Codex is not exposing a single undifferentiated quota, it is showing at least three distinct buckets in the product UI:

• A five-hour usage limit
• A weekly usage limit
• A separate code review allowance

That lines up with OpenAI's plan documentation, which describes local pairing, cloud task delegation, and automated GitHub code review as separate product surfaces. The reset therefore touched more than chatty local usage, it appears to have refreshed the broader Codex workbench that OpenAI is still trying to meter in real time.