Know what the numbers mean.
UsageReset is an independent community utility for public announcements, historical context and your own reset times. We are not affiliated with OpenAI or Anthropic.
Three different kinds of time
- Regular reset announcements: publication time is not the time every account is refilled.
- Banked credits: eligible users can redeem these later. A grant does not reset the regular-reset clock.
- Personal windows: your official account is the source of truth. Public announcements never overwrite your saved timer.
How public data arrives
Codex announcements come from the Codex Resets public API, with original text and source links retained. We load up to 100 recent announcements and show a 12-week calendar. Regular counts and median intervals exclude banked credits. Multiple announcements can refer to one reset; counts are not unique account refills.
The server rechecks about every two minutes and browsers read our cached response. Upstream failures may delay updates; previous successful records are retained when available, with their source timestamp. Calendar days use UTC; announcement times use your selected display timezone.
An estimate is not a promise
24/48-hour percentages prefer third-party estimates from codex-reset.com and have not been independently validated by UsageReset. Historical models and announcement signals use different assumptions and are not account telemetry. Forecasts older than 15 minutes are hidden. Historical intervals are not a schedule; waiting longer does not guarantee an imminent reset.
Read the upstream methodology and backtest ↗
Historical fallback when the forecast source is unavailable
When the upstream forecast is unavailable or stale, we use the mean m (in days) of up to eight recent completed regular-announcement intervals, and estimate a t-day probability as 1 − exp(−t / m). At least three intervals are required. Banked credits are excluded and identical timestamps are deduplicated. Separate announcements may still describe one reset, so this is an announcement-history baseline.
This model assumes independent arrivals at a constant rate. It is uncalibrated and has no demonstrated predictive advantage. The page labels it Historical baseline and shows its sample size. A longer current wait does not increase the probability; stale announcement data hides the estimate.
What Claude support includes
Claude support includes personal short/weekly timers, client JSON import and official service status. No reliable global-reset forecast feed is connected, so no speculative percentage is shown. Service health does not describe your personal quota.
Claude Status ↗ · Claude Code status line ↗ · Codex App Server ↗
Personal data and reminders
Saved times, usage percentages and imported quota fields stay in this browser’s localStorage. Raw JSON is never uploaded. Do not paste passwords or tokens. Browser notifications require permission and this page to remain open; background throttling, sleep and system settings may delay delivery. Calendar exports include a 15-minute advance alarm, subject to your calendar app’s settings.
A completed countdown only means the time you saved has arrived. It does not confirm a refill or automatically redeem a credit.
Back to reset radar →