Codex reset notices, history, and probability
Check Codex reset forecasts, official updates, and recent reset history in one place.
Current status
Reset history
The timing of regular resets may vary by user.
Past random reset times are grouped by hour on desktop and by two-hour blocks on mobile.
Reset records: n=…
Past intervals between consecutive random resets are grouped by elapsed time. Each interval runs from one random reset to the next.
Tibo (@thsottiaux)
Follow Samir for dot updates
Answers about Codex and ChatGPT Work usage limits and resets.
It is a statistical reference forecast indicating the likelihood of an upcoming reset. An empirical baseline derived from historical cadence is dynamically calibrated using official notices, OpenAI Status incident logs, and community signals. It is not an official probability from OpenAI.
Regular resets follow scheduled periodic refreshes. Special or compensation resets (often called wabi-resets or celebration resets) are unscheduled events executed by OpenAI following service outages, capacity issues, or milestone events.
Yes. On eligible plans, Codex and ChatGPT Work share the same agentic usage and credits pool. A reset affecting that shared pool restores limits for both Codex and ChatGPT Work, though applicability may vary depending on plan specifics.
Teaser posts are analyzed separately from official notices. When terms like tomorrow, tonight, weekdays, or specific hours are resolved, the forecast weighs the hinted window overlap and hint strength against the 24h and 48h windows. Unresolvable hints decay gradually over time.
A Banked Reset allows you to manually refresh your usage limits on demand. It is best used strategically: when reset likelihood is low and immediate progress is essential, use your Banked Reset; when likelihood is high or an official notice is imminent, consider waiting for a potential global reset.
The 24-hour and 48-hour values are unofficial statistical estimates based on past random-reset intervals and current observed signals. They are not probabilities published by OpenAI or a confirmed reset schedule.
We keep official notices, hints from Tibo and other people involved in development, public signals such as OpenAI Status, and resets confirmed as having happened distinct. Confirmed reset history is accumulated and compared with current conditions.
Voluntary support through GitHub Sponsors helps sustain the work of collecting reset information, organizing history, and improving forecasts. Recurring support at any amount helps with development and operation.