← Back to all posts

What to do when you hit your Claude Code or Codex usage limit

5 min readBy The Octuo team

You're halfway through a refactor and the agent stops: usage limit reached, try again later. It comes up constantly in the Claude Code and Codex issue trackers, and no trick makes it go away. The limits belong to the vendors. What you can control is how much work you lose when you hit one, and how often it happens.

Octuo demo video (90 seconds), also on YouTube.

How the limits work

Both services meter usage per account over time windows, not as a message count you can predict.

  • Claude Code on Claude's Pro and Max plans has a five-hour session limit and a weekly limit. They are shared with the Claude apps, so chatting on the web and coding in the terminal draw from the same allowance. You see warnings as you approach a limit.
  • Codex on ChatGPT plans shares usage with ChatGPT Work. OpenAI's docs describe five-hour periods on some plans and say weekly limits may also apply. How fast you use it depends on the model, the amount of context, tool use and whether the task runs locally or in the cloud.

The exact numbers vary by plan and change over time, so trust the vendor's page over any blog post, this one included.

The next ten minutes

  1. See where you stand. Run /status in Claude Code or in the Codex CLI to check your remaining allowance and when it resets. Codex also has a usage dashboard.
  2. Save the state of the work. Before you close anything, write down the goal, what is done, what is half-done and the next step. Commit the partial change to a branch. When the window resets, or another tool picks the task up, that note saves the next hour. Our restart note template is one way to do it.
  3. Know the vendor's options. Anthropic lists waiting for the reset, moving to a larger plan, turning on extra usage, or switching to API billing through the Claude Console. OpenAI lists buying additional credits on Plus and Pro, or running local chats with an API key at standard API rates. These are the vendors' options to define and price, so read their pages before you choose.
  4. Check you're not on the API by accident. Anthropic notes that if an ANTHROPIC_API_KEY environment variable is set, Claude Code uses that key instead of your subscription, and you're charged at API rates. Worth a look if your bill doesn't match your plan.

Make the allowance last longer

Most of the advice in both vendors' docs comes down to sending less:

  • Keep prompts precise and drop context the task doesn't need.
  • Point the agent at the relevant files, not the whole repo.
  • Keep CLAUDE.md or AGENTS.md short. It loads in every session.
  • Turn off MCP servers you aren't using. Each one adds context to every message.
  • Split big jobs into smaller tasks with a clear finish line, so a limit interrupts one piece instead of everything.

What we wouldn't do: open extra accounts to get around a limit, or borrow someone else's login. Each service's terms decide what's allowed.

What Octuo changes, and what it doesn't

Octuo is an assistant on your Mac that works through your other tools, including the Claude Code and Codex CLIs signed in on your Mac.

It doesn't change the limits. Each account's usage counts against that account, under that service's terms. Octuo doesn't raise, reset or pool limits, it doesn't move allowance between accounts, and we don't promise it saves you money.

What it can change is what happens around a limit:

  • The hand-off. If one tool is out for now, Octuo can give the next piece to another tool you have, or do it with one of the models it works with directly, and it passes along the context so you don't explain the task again. Work done on Octuo's own models is Octuo usage, billed separately from your subscriptions.
  • The waiting. Long, multi-step jobs keep going after the chat closes. A scheduled routine can pick a job back up later, for example after a reset, and Octuo tells you in Attention when it's done or needs your decision.
  • The bookkeeping. Results from every run come back to one place, so a limit in the middle of a job doesn't leave it scattered across windows.

If that is how you'd like limits to be handled, see what Octuo does.

FAQ

Does Octuo give me more Claude Code or Codex usage? No. Anthropic and OpenAI set the limits, and they stay the same.

Does Octuo combine my accounts' limits? No. If you've connected more than one account you're authorized to use, Octuo can hand work to each, but each one's usage counts against that account under its service's terms. We don't suggest adding accounts to get around a limit.

Is it fine to keep going with an API key after the limit? Both vendors offer it, billed at API rates. Make that choice on purpose; don't let a leftover environment variable make it for you.

Will a running task be cut off at the limit? OpenAI says an active Codex turn can continue when you reach your limit, subject to fair use. Anthropic shows warnings as you approach a limit. Either way, save your state when you see a warning.

Sources

The product names above belong to their owners. We have no affiliation with them.