Claude Code Usage Limits Explained: Why You Hit Them and What to Do
Published 2026-11-11
If you use Claude Code on a subscription, sooner or later you may see a message that you have reached your usage limit and need to wait. This article explains in general terms why that happens, what it looks like, and what you can do about it, including the option of paying per token through an API gateway. We do not quote exact limits here. Anthropic defines them, changes them, and they differ by plan, so the authoritative source is Anthropic's own help center and plan pages.
Disclosure: we run OriginCoder, an API gateway, so the gateway option in this article is our product. We try to state the trade-offs fairly, and the official subscription is the better choice for many people. This article is current as of November 2026.
Why limits exist
A subscription has a fixed monthly price, but each request to a large model costs real compute, and a coding agent can send many large requests: it reads files, plans, edits, runs tools and loops. To keep a flat price sustainable, providers cap how much a plan can use over a period of time. How the cap is measured (messages, tokens, compute, sessions), how long the window is, and whether there is also a longer weekly or monthly cap are details set by Anthropic. Check the current description for your plan.
What it usually looks like
- A notice in Claude Code that you have reached the limit, usually with a time when it resets.
- Being unable to continue a long session until the window resets.
- Hitting the limit faster than you expected on large repositories, long conversations or heavy agentic tasks, because each turn carries a lot of context.
- Differences between plans: higher tiers raise the limits, but do not remove them.
These are general patterns, not a statement of how any specific plan counts usage. If a number matters to you, look it up in the official documentation.
What uses your allowance fastest
- Very long conversations. The whole context is processed again on each turn, so long sessions cost more per message than short ones.
- Large files and wide searches pulled into context.
- The most capable model on every task, including simple ones.
- Long autonomous runs with many tool calls.
- Several sessions running at the same time.
Ways to stretch a subscription
- Start fresh sessions for new tasks. Claude Code has commands to clear the conversation (
/clear) and to summarise it (/compact). Check the current docs for exact behaviour. - Give it tighter context. Point to the specific files and directories, rather than letting it read everything.
- Use a lighter model for simple work, and the strongest one for hard problems.
- Plan before you run. A clear prompt avoids many wasted loops.
- Look at your plan's usage view to see what is consuming the allowance.
When you still run out: your options
| Option | How it works | Good when | Trade-offs |
|---|---|---|---|
| Wait for the reset | Do other work until the window renews | Occasional, short interruptions | Blocks you when you are in flow |
| Move to a higher plan | Higher tier, higher limits | You hit limits regularly and use Claude Code daily | Higher fixed fee; limits still exist |
| Anthropic API key | Pay per token directly to Anthropic through the console | You want the official vendor and metered billing | No flat-fee cap on cost, so set a budget |
| API gateway, pay as you go | Point Claude Code at a gateway with a Base URL and key, pay per token | Bursty use, several providers, per-request cost records | Third party in the path, may lag new features, your own cost control |
| Another tool or model for overflow | Use a different agent or model when Claude is capped | You can work with a second tool | Different quality and workflow |
Pay-as-you-go has no ceiling that stops you, which is both its advantage (you are not locked out) and its risk (a long agentic session can cost more than you expect). Check per-model prices before you start, and look at your usage records after the first sessions.
Setting up the gateway option
With OriginCoder, you set an environment variable for the address and one for the key, and claude runs as before. The steps are in How to use Claude Code with a custom API endpoint. Per-model prices are on the model pricing page. For a fuller comparison of the two routes, including who should stay on the subscription, read Claude Code alternative: official subscription vs. API gateway.
Things to check before switching
- Terms. Using any plan, API or gateway means accepting that provider's terms. Make sure your use complies with them.
- Privacy. A gateway sees your prompts and code context. Read its terms and avoid sending secrets.
- Compatibility. Test your normal workflow on a small task first.
- Keep the subscription if it fits. If the limits rarely bother you, there is no reason to move. A gateway can complement a subscription instead of replacing it.
OriginCoder is an independent platform and is not affiliated with Anthropic.