Claude Code Usage Limits Explained: How to Use Fewer Tokens

In Claude Code, usage means the tokens your sessions consume: all the text Claude reads from your files and conversation, and all the text it writes back. On Pro, Max and Team plans that usage comes out of an allowance that resets on a rolling five-hour window, with a weekly cap on top, and it is shared with your Claude chats. On an API account you pay for every token instead, and Anthropic doesn't publish a fixed number of messages for any plan.

The rest of this guide explains what counts, how each plan measures it, how to check where you stand, and eight habits that stretch the same plan much further.

What counts as usage in Claude Code

Every time you send a message, Claude Code sends the whole conversation again: the instructions, your CLAUDE.md, every earlier message and every file it has read. That is why, as Anthropic's cost guide explains, a one-line question late in a long session draws on your usage for the entire conversation.

Beyond your messages, these add to the total:

• Files and outputs Claude reads. Big logs, spreadsheets, PDFs and images all go into the context, and stay there.
• Thinking. Extended thinking is billed as output tokens, and a higher effort level means more of it.
• Helpers. Each subagent or workflow that Claude starts sends its own requests on top of yours.
• Scheduled work. A /loop or scheduled task fires on its interval even while you are away, re-sending the context each time.
• Small background requests. Features such as conversation summaries for resuming cost a little even when idle, which the docs put at typically under $0.04 per session.

How plan limits work

Anthropic's pricing page says every plan has limits that reset on a rolling five-hour session window, paid plans add weekly limits, and your activity on the web, desktop, mobile and in Claude Code all draws from the same pool.

How to see your usage

On paid plans, the /usage breakdown attributes recent usage to skills, subagents, plugins and individual MCP servers, and it flags behaviors such as long context or cache misses once they cause a tenth of your recent usage. Before a window runs out, Claude Code can also warn you with a line such as You've used 85% of your session limit · resets 3:45pm.

Why you hit limits sooner than expected

• One long session for everything. Every message carries the full history, so unrelated tasks in the same session keep paying for each other.
• Coming back after a break. On a subscription the prompt cache lasts an hour, dropping to five minutes once you draw on usage credits, so the first message after a long break re-processes the whole context without the cache discount.
• The biggest model at high effort for small jobs. Renaming files doesn't need the most capable model thinking hard.
• Many helpers at once. Subagents and workflows multiply requests, and Anthropic's docs say agent teams use roughly 7 times more tokens than a standard session when teammates run in plan mode.
• Loops and schedules left running. They keep firing while the session sits idle.
• Heavy chat use the same day. Your Claude chats come out of the same pool.
• Big bursts. Usage counts against the session and weekly allowances at the same time, so one heavy burst can use up the weekly limit before the five-hour window resets.

What happens when you hit a limit

The limit messages are listed in Anthropic's error reference, and each one names its reset time. In an interactive session signed in with a subscription, Claude Code can wait and continue the task on its own after the reset, as long as you leave the session open.

You also have these choices:

• /rate-limit-options shows ways to keep going, including waiting and continuing automatically.
• /usage-credits turns on usage credits, which keep you working past your plan's limit at standard API rates; on Pro and Max you can set a monthly spend limit in Settings > Usage.
• /upgrade opens the page to move to a higher plan for bigger base limits.
• A Claude Console account with API credits, for an intensive sprint billed per token.

8 ways to use fewer tokens

Model choice is the easiest lever to see in numbers: on the API, output prices differ by a factor of 100 between the smallest and the largest current model.

1. Clear between tasks. /clear starts a fresh conversation and costs nothing, so old context stops riding along on every message. Run /rename first if you want to come back to it with /resume.
2. Compact at natural breaks, with a focus. /compact keep the decisions and the list of files we changed shrinks the history while keeping what matters. Do it between tasks rather than mid-task, and use the rewind menu's "Summarize from here" to compress only part of a session.
3. Match the model to the job. Opus 5.5 is the default on paid plans; switch with /model to Sonnet for everyday work and to Haiku for simple, mechanical jobs.
4. Lower the effort for simple work. /effort low suits renames, formatting and quick edits, and Opus 5.5 already defaults to medium. For one hard question, type ultrathink in that prompt instead of raising the effort for the whole session.
5. Be specific and point at files. "Fix the price table in @index.html" reads one file; "improve the site" makes Claude scan everything.
6. Keep CLAUDE.md lean. It loads into every session, so keep it under 200 lines and move step-by-step procedures into skills, which only load when used. The CLAUDE.md guide shows what belongs where, and how to create a Claude skill covers the rest.
7. Switch off what you don't use. Disable idle MCP servers in /mcp, ask for summaries instead of full log dumps, and hand log-heavy work to a subagent so only its summary comes back.
8. Plan big changes and stop early. Plan mode prevents expensive rework, Esc stops a wrong turn immediately, and after two failed corrections a /clear with a better prompt beats a third attempt.

When a long task has to continue tomorrow, a handoff note lets you start fresh without losing the thread:

Write a short handoff note to handoff.md: what we were working on, what's finished, what's left, the files involved and the decisions I made. Keep it under 20 lines.

Then run /clear, and begin the next session with "Read @handoff.md and continue from there". The Claude Code cheat sheet lists every command mentioned here on one page.

What to do next

Open /usage in your next session and look at the breakdown before you change anything, so you know whether long context, helpers or a heavy model is your biggest drain. Then pick two habits from the list, such as clearing between tasks and lowering effort for routine work, and use them for a week. The beginner tutorial builds these habits into a first project, and the Claude Code course has lessons on prompting without wasting time and on giving Claude the right context.