How Codex and Claude usage-limit resets actually work (and how I track them)
If you use Claude Code or Codex every day, you've probably had this moment: you hit your limit, then someone on X says "limits just got reset" and you have no idea if that applies to you, if it already happened, or if it's still rolling out. I kept refreshing X to figure this out, so I built a small tracker with Claude Code that logs every reset with its source post. After a few months of…
When using Claude Code or Codex daily, users often experience hitting their usage limit. This can lead to confusion about when the next reset will occur. To help track these resets, a small tracker was built using Claude Code. The tracker logs every reset with its source post.
There are two types of resets: normal windows and extra, one-off resets. Normal windows include a rolling short window, which resets every five hours for Claude and five hours for Codex, and a weekly window that resets every week for both products. Codex Plus and Business Standard users also have a weekly allowance that must be left in order to continue using the service. Both products have fixed reset days and times, which can be found in Settings > Usage.
Extra, one-off resets are less predictable. These are announced by Anthropic or OpenAI on X, and users must wait for the reset to be applied to their account. The tracker shows the median gap between these resets, but the exact timing is unknown. Banked resets are another type of reset that can be redeemed at the user's discretion. These are credits that users can use when they choose, but they require certain conditions to be met before they can be used.
The tracker provides a way to monitor resets and stay informed about when they occur. Users can view the median gap between resets, the number of resets in the last 7, 30, and 90 days, and even set up alerts using RSS feeds or a calendar subscription. This allows users to better manage their usage and plan for resets accordingly.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
