Home › Limits
Coding agent usage limits, explained
Where to see your own meter, what happens when it runs out mid-task, and what Crossing Guard does about it.
Where to see your meter
From each vendor's documentation, read on 2026-10-04. Plans change often, so follow the links for current numbers.
| Agent | Check it | How the limit works |
|---|---|---|
| Claude Code | /usage shows your plan usage bars and what counted against them. | A limit message names a session limit or a weekly limit and says when the window resets. Team and Enterprise seats reset on a rolling five-hour window and a weekly window. Anthropic's documentation → |
| Codex | /status in the CLI, or the usage dashboard in your ChatGPT settings. | Allowances are given per five-hour period on some plans, and weekly limits may also apply. OpenAI's documentation → |
| OpenCode | Check the provider you connected. | OpenCode calls the provider you configure, so the limit is that provider's. |
The console's Usage view adds up the token counts your agents already record in their local session files: input, output and cache hit rate, per agent and per day. It reads those files and estimates nothing. It does not know your plan's limit. Source →
When you hit the limit mid-task
Claude Code shows when the window resets, and recent versions can wait and continue the interrupted task after the reset. Codex lets the agent finish the turn it is on. Both vendors sell extra credits.
If you would rather keep going in another agent, Crossing Guard gives the next one three things:
The same memory
Claude Code and Codex read one store of Markdown records at session start. Source →
The same rules
One rule document is evaluated for every connected agent.
A handoff note
The console drafts it from the stalled session. You edit it, and it is written into the repository for the next agent to read. Source →
The session itself does not move. The next agent starts a new session with that context.
Helper and reviewer runs
Agents that Crossing Guard starts itself (helpers, reviewers and followers) are handled differently. When the provider answers with a rate limit, a quota or usage limit, an outage or a sign-in error, the run is not counted as failed. It is tried again on the same route up to three times, then moved to the next route in the list you wrote, and then parked for you if no route works. A route names an agent and a model, so the fallback can be a different vendor. Source →
We have tested this against a simulated provider, and not yet through a real vendor outage.
Asked constantly
Is Claude Code free?
No. It runs on a paid Claude plan (Pro, Max, Team or Enterprise) or on per-token billing through the API or a cloud provider. See Anthropic's pricing.
Is Codex free?
OpenAI lists Codex as included in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise plans, with a different allowance on each. See OpenAI's pricing.
Does switching models get me past the limit?
On Claude Code, a session or weekly limit is shared across all models, so switching does not help. After a limit that names one model family, switching to a model outside that family does.
Is Claude or Codex down?
Check Anthropic's status page or OpenAI's status page. If only your own session is stuck, it is more likely a limit than an outage.