Codex rate limit reached? Your options, from waiting to an API key
Your current turn still finishes. After that you can wait, buy credits, switch to an API key, or set up an alert so it does not surprise you again.
By ClaudeGPT Si Newsroom, Editorial team
3 min read
Codex rate limit reached means you have used your plan's allowance for the current window, but the turn you were in will still finish. From there you can wait for the reset, buy credits on Plus or Pro, or switch to an API key, and next time a desktop tracker like UsageAtlas can warn you before you get that far.
Here is the order we would work through it.
1. Let the current turn finish
OpenAI says that if you reach the limit during an active turn, the agent can keep working on that turn, subject to fair use. Do not kill the session in a panic. Let it complete, then commit or stash what it produced.
2. Check what you hit and when it clears
Run /status in the CLI. OpenAI's reference says it shows the chat ID, context usage and rate limits. For reset times, open the Codex usage dashboard at chatgpt.com/codex/settings/usage; if you signed in with ChatGPT, /usage takes you to usage and billing settings.
It matters which window you hit. On Plus and Standard Business the allowance is estimated per five-hour period, and OpenAI notes weekly limits may also apply. A five-hour stop is a coffee break. A weekly stop can cost you days.
3. Pick a way to keep working
| Option | Who can use it | Trade-off |
|---|---|---|
| Wait for the reset | Everyone | Free, but you are blocked until then |
| Buy credits | Plus and Pro | Extra spend, no plan change |
| Workspace credits | Business, Edu, Enterprise on flexible pricing | Billed to the workspace |
| API key for local chats | Any user | Charged at standard API rates, no cloud features |
API token prices are separate from subscription usage. If you switch to a key, watch that bill on its own.
4. Find out why it happened
OpenAI does not publish fixed message limits because the cost of a message varies. The things that push it up:
- the model you picked
- long, multi-step tasks
- sessions holding a lot of context
- running tasks in the cloud rather than locally
- reasoning, tool use and retrieval
Local messages and cloud chats share one allowance, so a cloud task left running can eat into the budget for your terminal work. Splitting work into smaller sessions and picking a lighter model for routine edits are the usual fixes.
5. Set up a warning for next time
The CLI only tells you when you ask. That is the gap a tracker fills.
UsageAtlas is the tool we recommend here. It is a free, open-source desktop app for Windows, macOS and Linux that reads Codex usage through the signed-in Codex CLI on your machine. It shows session and weekly windows, remaining capacity, reset times, and plan and credits where Codex reports them. You set a threshold and it sends one desktop alert per window, so you can wrap up a task before the wall rather than at it.
It also keeps a daily history of tokens, models and estimated cost, which is how you spot the one kind of task that keeps draining the week. No account is needed for local tracking, and it does not read prompts or responses. Cost figures are estimates, not your bill.
If you rarely hit limits, skip the extra app: /status and the dashboard are enough. The tracker pays off when limits interrupt real work more than once a week, or when you also run Claude Code or Cursor and want every tool's limits in one window.
Quick checklist
- Let the turn finish.
- Run
/status, then check the dashboard for the reset time. - Wait, buy credits or use an API key.
- Trim the heavy tasks.
- Add an alert so the next one is not a surprise.