When Codex stops at a usage limit, first identify which account and which limit stopped the request. A ChatGPT allowance, an API balance, a spending cap, and a request-rate limit need different fixes. Find the matching row below before buying credits or changing configuration, so your next action restores the workflow you intended.
Match the message to the right limit
| What you see | Where to check | Useful next action |
|---|---|---|
| Plan usage exhausted or a reset time | Codex usage dashboard and active account | Wait for the relevant reset or review that account's additional-usage options |
| Credit balance exhausted | The API provider's billing dashboard | Fund the account actually receiving the requests |
| Project or organization spend cap reached | That project's spending controls | Ask the budget owner to review the cap |
| Requests or tokens arriving too quickly | Error details and retry guidance | Reduce concurrency and wait before retrying |
| 401 or invalid authentication | Login method, provider, and credential source | Follow the authentication branch instead of adding credit |
OpenAI's API error reference separates balance, spend, rate, and authentication errors. Read the actual message as well as the HTTP status: several different failures can share a status code. A custom provider may use its own error names.
Check the account before the counter
# In your terminal: identify the CLI and login method
codex --version
codex login status
# Inside an active Codex CLI session:
/statusFor ChatGPT access, open the Codex usage dashboard and compare its account with your CLI login. OpenAI's pricing documentation explains that local and cloud work share plan usage, and that task size and model choice affect consumption. Five-hour estimates are not a guaranteed message count; weekly limits may also apply.
Record the reset time and the blocked action. For example, “I can start a small local task, but this cloud task is unavailable” is more useful than “Codex is out of tokens.” It tells you what needs restoring and avoids purchasing an option that serves a different workflow. In a managed workspace, include the workspace name when checking access with your administrator.
Choose how to continue
- If the deadline can wait: keep the current diff and a short note of the next step, then resume after the displayed reset.
- If the plan fits but today is unusually busy: check additional credits. OpenAI documents this option for Plus and Pro, with separate workspace arrangements for eligible business plans.
- If this happens repeatedly: compare the cost of a higher allowance with your actual recent workload before upgrading.
- If you want usage billing for local work: evaluate the API route separately, including model access, rates, and required features.
The API route has its own balance and limits. Keep enough room in your trial budget for a failed attempt and a correction; a single successful response does not tell you the cost of finishing the task. The Codex pricing guide helps you compare billing shapes without treating an API key as unlimited usage.
Handle an API rate limit without a retry storm
Stop parallel retries first. When the provider supplies a retry delay, use it; otherwise space attempts out and bound the retry loop. OpenAI's rate-limit guide recommends backoff and warns that unsuccessful requests can still count against the minute limit. A billing error requires a billing action, so endlessly retrying it only obscures the original problem.
Then inspect the workload: several terminals, a background script, and an IDE extension may all be using the same API project. Pause one source at a time. A controlled single request gives you a clearer diagnosis than changing the model, key, and concurrency together.
Make the next session easier to budget
Define the change and its acceptance check before starting. Point Codex to the relevant files, keep unrelated work in a separate task, and review a checkpoint before asking for optional improvements. Track the accepted result and total usage together. This gives you a practical baseline for deciding whether a plan or API spending works better for your projects.
To evaluate Kunavo for local Codex CLI work, follow the Responses-provider configuration after creating an account. Keep this as a separate provider trial: its balance does not replenish ChatGPT usage or unlock OpenAI cloud features. For a rejected login, use the Codex 401 troubleshooting guide.
FAQ
When do Codex usage limits reset?
Check the reset times shown for your account in the Codex usage dashboard or /status. A shorter usage window and a weekly limit can apply together. A reset in one window does not mean every limit has reset.
Does a new API key reset my Codex limit?
No. Rotating credentials does not replenish a ChatGPT plan allowance or increase an API project's limits. Choosing separately billed API access changes the billing route for supported local work; it does not reset the original account.
Why do I hit the limit after fewer messages than another user?
A message is not a fixed unit of work. Models, context, tool activity, and task complexity affect usage. Compare the same kind of task and check the account's usage display instead of treating someone else's message count as an allowance.
Will adding API credit fix a 401 error?
A 401 needs authentication diagnosis first. Confirm the account or provider, credential source, and destination host. Adding balance does not correct a rejected credential. A balance or spending error needs its own billing remedy.
Official documentation checked September 17, 2026. Your account dashboard supplies the current allowance and reset times.