Codex "You've hit your usage limit": what it says and what to do
What each ChatGPT plan's Codex usage-limit message means, why the reset time is off on a UTC server, and what codex exec does when it hits the limit.
“You’ve hit your usage limit.” means your ChatGPT plan’s Codex allowance is spent for now, and OpenAI refused the request until it resets. The rest of the sentence depends on your plan: Plus is offered Pro or credits, Pro credits, Free and Go an upgrade to Plus, Business a request to the admin. The time after “try again at” is in the machine’s own timezone, so on a server set to UTC it is not your local time. Wait for that time, buy credits if your plan offers them, or run the job on an API key, billed separately. A codex exec job does not retry it: the run fails and exits with status 1.
What each plan’s message says
The sentence is assembled in Codex’s error definitions from the plan OpenAI reports back with the refusal (as of release 0.169.0, November 2026). With a reset time of 3:42 PM, they read:
Plus: You’ve hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/settings/usage to purchase more credits or try again at 3:42 PM.
Pro: You’ve hit your usage limit. Visit https://chatgpt.com/settings/usage to purchase more credits or try again at 3:42 PM.
Free, Go: You’ve hit your usage limit. Upgrade to Plus to continue using Codex (https://chatgpt.com/explore/plus), or try again at 3:42 PM.
Business: You’ve hit your usage limit. To get more access now, send a request to your admin or try again at 3:42 PM.
Enterprise, Edu, or plan not known:
You’ve hit your usage limit. Try again at 3:42 PM.CopiedThe Pro line covers all three Pro plans; the Business line covers Team, the Business seats and some Enterprise plans. When OpenAI sends no reset time, “try again later” takes the place of the time. Two other forms replace the plan sentence:
- “You’ve hit your usage limit for <limit>. Switch to another model now, or try again at …” One model’s own limit is spent, not your whole allowance.
/modelin the CLI, or-mon acodex execrun, picks another. - A promotional message from OpenAI after “You’ve hit your usage limit.”, when OpenAI sends one with the refusal.
A workspace that has run out of credits or reached its spend cap gets one of four sentences instead, depending on whether you own the workspace:
- “Your workspace is out of credits. Add credits to continue.”
- “Your workspace is out of credits. Ask your workspace owner to refill in order to continue.”
- “You hit your spend cap set in your workspace. Increase your spend cap to continue.”
- “You hit your spend cap set by the owner of your workspace. Ask an owner to increase your spend cap to continue.”
The spend cap is the workspace owner’s setting, and only an owner can raise it or add credits.
Messages that look like it and are not
- “Selected model is at capacity. Please try a different model.” OpenAI had no room to run the model; your allowance is untouched. The capacity error’s guide covers it.
- “Quota exceeded. Check your plan and billing details.” This is the API’s refusal when Codex runs on an API key whose account has no credit left, not a ChatGPT plan limit. It is settled in the OpenAI Platform account’s billing.
- “To use Codex with your ChatGPT plan, upgrade to Plus: https://chatgpt.com/explore/plus.” The plan does not include what was asked for, which no reset will change.
The reset time on a server
OpenAI sends the reset as a moment in time, and Codex converts it to the machine’s local timezone before printing it, with no zone name after it. A time on today’s date prints as 3:42 PM; a later day prints as Nov 17th, 2026 3:42 PM. Many cloud servers run on UTC, so a server’s “try again at 3:42 PM” is 3:42 PM UTC, which in New York is 10:42 AM. To see which zone the machine uses:
timedatectl | grep 'Time zone'
dateCopieddate prints the machine’s current time in that same zone, so the gap between it and the time in the message is how long the job has to wait.
What a codex exec job does when it meets it
Codex never retries this error by itself: in its source a usage limit is in the group of errors with no retry, unlike a dropped stream or a timeout. In codex exec, that means the message on standard error, an error event if the run uses --json, and, by its source, an exit status of 1, the same as any other failed run. OpenAI’s Codex pricing page says that when a limit is reached mid-task, “the agent will be able to continue working on that turn, subject to fair use limits”, so it is usually the next request or the next run that is refused.
Retrying in a loop does no good before the reset, and a job that runs every few minutes simply fails every few minutes until then. A wrapper is better off telling this failure from the others and saying so:
#!/bin/sh
# Run a codex exec job; on a usage limit, stop and report the reset time.
out=$(codex exec "$1" 2>&1)
status=$?
if [ $status -ne 0 ] && printf '%s' "$out" | grep -q 'hit your usage limit'; then
printf '%s\n' "$out" | grep 'usage limit' | mail -s "codex: usage limit" you@example.com
fi
exit $statusCopiedReplace the address and the mail command with whatever reaches you. Rerun a job that got partway with codex exec resume --last after the reset rather than from the start, as the capacity guide describes. A scheduled job also shares the allowance with your own sessions on the same account, so a busy day at the keyboard can leave the night’s job nothing; the same is true of Claude Code, and running Codex and Claude Code on a schedule covers the rest of an unattended job.
How much is left
OpenAI’s pricing page names two places: the Codex usage dashboard, and /status inside a Codex CLI session, “to see remaining limits”. The five-hour window is not the only one: “Weekly limits may also apply,” and a weekly reset can be days away, which is when the message prints a date. How large each plan’s allowance is, and which plans have no five-hour window, is in which ChatGPT plans include Codex.
The choices
- Wait for the reset. Nothing on the machine has to change; the next request after the time in the message goes through.
- Buy credits. On the pricing page, Plus and Pro users “can purchase additional credits to continue working without needing to upgrade their existing plan,” and Business, Edu and Enterprise plans with flexible pricing “can purchase additional workspace credits to continue working.” Credits are bought from the usage page the message links. On a Business workspace, the admin or owner decides.
- A bigger plan. Free and Go are pointed to Plus, and Plus to Pro; the plans guide compares what each gets.
- An API key. “All users may also run extra local chats using an API key, with usage charged at standard API rates.” That bill comes from your OpenAI Platform account, not your ChatGPT plan, so it is the usual choice for a job that must run on time. Give a single run its key with
CODEX_API_KEY(anOPENAI_API_KEYin the environment is not used bycodex exec, as the 401 guide explains):
CODEX_API_KEY=<api-key> codex exec "triage open bug reports"CopiedOn an API key the allowance is your Platform account’s credit, and running out of it prints “Quota exceeded” rather than this message.