How do usage limits work?
Usage limits are caps on how much you can use a subscription-based AI tool within a given period. Claude and Codex subscriptions combine a 5-hour usage window with weekly limits.
The 5-hour window and weekly limits
Claude plans count usage in 5-hour windows; once a window is used up, you wait for the next. On top of that sit weekly limits, so with heavy use the weekly cap can stop you even after the 5-hour window resets. Codex used through ChatGPT plans also has 5-hour and weekly limits. The exact size depends on the plan, and providers adjust them from time to time.
What burns through limits
Limits aren’t published as exact token counts, but they fill up with how much you use.
- Larger models: the same job takes a bigger bite out of the limit on a stronger model.
- Long context: every turn resends the whole conversation.
- Parallel agents: every agent on the same account draws from the same limit.
- Cache misses: reprocessing a conversation after long pauses.
How it differs from API rate limits
With the API you pay per token; the limits there are rate limits, like requests and tokens per minute, and spending shows up directly on your bill. On a subscription you pay a flat fee and usage limits cap what you can do. Most coding agents work with either.
Staying under the limit
Limits tend to run out at the busiest moment; a few habits push that back.
- Give simple jobs to a smaller model.
- Keep context small and turn off MCP servers you don’t need.
- Shorten long command output.
- Watch your remaining usage and wrap up big jobs before the limit runs out.
Usage limits in AgentVera
AgentVera makes limits visible and offers account chains so work doesn’t stop when one runs out.
- The status bar shows each account’s 5-hour and weekly usage.
- Automatic account switching: put Claude Code and Codex accounts in a chain; when the remaining limit drops below your threshold or the CLI reports a limit, the agent moves to the next account and the conversation is copied over (Pro and Team).
- Telegram notifications can warn you about limits on your phone.
- Automatic model gives small jobs to Haiku and saves Opus for hard ones.
- Automatic account switchingLine up Claude and Codex accounts; when one hits its limit, the agent moves to the next.
- Automatic modelEach task is rated simple, normal or hard, and the agent switches to that tier’s model.
- Token savingsContext profiles, shorter command output with RTK, live context size and automatic /compact.
FAQ
When does the Claude Code limit reset?
The short-term limit renews in 5-hour windows and the weekly limit on a weekly cycle. Claude’s own interface shows your remaining usage and when it resets.
Do Claude and Codex share limits?
No. Claude is tied to an Anthropic account and Codex to an OpenAI account, and each has its own limits. Within one Claude account, though, the Claude apps and Claude Code draw from the same limit.
What can I do when I hit the limit?
Wait for the window to renew, upgrade your plan, or switch to an API key and pay for what you use.
Do parallel agents use up limits faster?
Yes. Every agent on the same account draws from the same limit, so giving small jobs to a cheaper model and keeping context small matters even more.