Codex 5-hour usage limit

What the five-hour window actually measures

Some ChatGPT plans provide Codex usage through both a shorter five-hour window and a longer weekly window. The shorter window controls how much included work can be done during that period. It is not a promise that every request costs the same amount, and it should not be converted into a universal message count. OpenAI says consumption varies with the model, task complexity, context, reasoning setting, tools, execution location, and speed. A small explanation or narrow code edit may therefore use much less allowance than a long agentic task that reads a repository, runs commands, and iterates on failures.

When the window starts and resets

The label “five-hour limit” does not necessarily mean a clock that resets for everyone at the same wall-clock time. OpenAI describes a new window as starting when you send your first Work or Codex request after the previous window ends. Your own Usage screen shows the relevant reset time for the active account and workspace. Treat that timestamp as authoritative; do not infer it from another user’s screenshot, subscription tier, region, or a public reset announcement.

How to check the correct allowance

First identify which meter is exhausted. A plan can have capacity left in its weekly allowance while the five-hour allowance is empty, or the reverse. Work and Codex may also share an allowance and credit pool when those features are available on your plan. Checking only one percentage can therefore lead to the wrong diagnosis.

1. Open ChatGPT or Codex and confirm the active account and workspace.
2. Open Settings → Usage and locate the five-hour and weekly meters.
3. Record the reset date, time, and timezone shown beside the exhausted meter.
4. In an active Codex CLI session, use /status as an additional view of current usage.
5. Paste the displayed time into the Reset Notes calculator instead of estimating it manually.

Why two tasks consume different amounts

Usage is workload-based rather than a flat charge per message. The same prompt can cost more when it carries a large conversation history, many attached files, a broad repository scan, repeated tool calls, or a high reasoning setting. Model choice also matters. This does not mean that users should always select the smallest model; it means the model and scope should match the job. A focused request with clear acceptance criteria is easier to execute and usually avoids spending allowance on exploratory work that the user did not need.

What to do when the five-hour allowance is empty

Start with the choices shown in your own Usage screen. Depending on plan, account, billing country, and availability, the interface may offer waiting for the displayed reset, using an available banked reset, adding usage credits, purchasing an eligible instant reset, or changing plans. These choices are not interchangeable. Credits are a pay-as-you-go balance for supported usage after included allowance is consumed. A banked reset is a saved promotional reset. A purchased instant reset applies immediately and can change the weekly schedule. Do not buy or redeem anything until you have checked which allowance is exhausted and what the confirmation screen says it will change.

Plan the next run without wasting the new window

Use the waiting period to narrow the next task. Separate must-have work from optional cleanup, link only the files that matter, and define a concrete completion check. If a task is large, split discovery, implementation, and verification into clear phases so a failed experiment does not drag the full repository context through every turn. Preserve useful decisions in project instructions or a short handoff rather than repeatedly pasting an entire conversation. These habits improve predictability, but they do not bypass or alter the account’s actual limits.

If the reset time arrives but access does not return

Refresh or reopen Settings → Usage and make sure you are still viewing the intended account and workspace. A dashboard can take a short time to update. If the displayed time has passed and usage remains unavailable, record the exact notice, the reset time and timezone, the client and model you were using, and a screenshot of the Usage screen. Contact OpenAI Support with those details. Support can investigate suspected account or purchase problems, but OpenAI says normal limits are not manually reset as a courtesy.