Spend less of your Claude Code quota

Claude Code charges by tokens, and on a subscription those tokens count against your plan limits. The more context Claude has to read, the more it uses. Most of the savings come from keeping that context small.

First, see where it goes

/usage shows the current session's token use. On a Pro, Max, Team or Enterprise plan it also shows what recent usage went to skills, subagents, plugins and MCP servers, with suggested changes. The dollar figure is for API users and is only an estimate.
/context shows what is taking space in the conversation.
The number beside the spinner is a live, approximate count of the output of the current turn. It does not include what you sent in.

Why a long session costs more than it looks

Claude Code sends the whole conversation with each request. Caching makes the repeated part cheaper, but a one-line question at the end of a day-long session still carries all of it.
If you leave a session idle for longer than the cache lasts, your next message reprocesses the full context.
A scheduled task or loop keeps firing in an idle session, and each run sends your full context. So does a message from another of your sessions.
Every subagent sends its own requests on top of the main conversation.
/compact has to read what it summarizes, so compacting a huge context is itself a big request. /clear costs nothing.

The habits

Clear between tasks/clear when you switch to unrelated work. Stale context is paid for on every later message. Use /rename first if you want to find the session again with /resume.
Steer the summary/compact Focus on code samples and API usage tells Claude what to keep. You can also add a "Compact instructions" section to CLAUDE.md.
Match the model to the jobThe documentation says Sonnet handles most coding tasks well and costs less than Opus, which it suggests keeping for hard design or multi-step reasoning. Switch with /model. For simple subagents, set model: haiku.
Lower the effortExtended thinking is on by default and its tokens are billed as output. For simple work lower the level with /effort. Some models always think.
Keep CLAUDE.md shortIt loads every session. Aim for under 200 lines and move workflow-specific instructions into skills, which load only when used.
Trim MCPTool definitions are deferred by default, so idle servers cost little. Command-line tools such as gh add no per-tool listing at all, and /mcp lets you disable servers you are not using.
Filter before Claude readsA hook can cut a 10,000-line log down to the error lines before they reach the conversation. See three hooks to paste.
Send noisy jobs to a subagentTest runs and log reading stay in its context and only a summary returns. Its requests still draw on your usage, so give it a smaller model. See write your first subagent.
Ask specific things"Improve this codebase" makes Claude scan broadly. "Add input validation to the login function in auth.ts" does not.
Plan first on big tasksShift+Tab switches to plan mode, where Claude explores and proposes before it edits, which avoids expensive rework if the first direction was wrong.
Stop earlyEscape stops a wrong turn at once. /rewind or a double Escape restores the conversation and code to an earlier checkpoint.
Give it a way to checkTest cases or the expected output in the prompt let Claude catch its own mistakes. Test one file at a time.
Stop what you no longer needA forgotten loop keeps spending. See how to stop an agent from looping.

Background use

Session recaps, idle compaction and prompt suggestions can also use tokens while you are away. The documentation says where to turn each one off.

A number, with context

Anthropic says that across enterprise deployments the average cost is around $13 per developer per active day, and below $30 per active day for 90% of users. That is API billing for companies, not a forecast for your plan, and it varies with model, codebase and how many instances you run.

Limits

We have not measured how much any of these saves. Anthropic's page is the authority, and commands, defaults and plan limits change.
Smaller models and lower effort can give weaker answers on hard problems. Try one change at a time.
None of this is a guarantee about your bill or your limits.

This follows Anthropic's "Manage costs effectively" page as we read it in October 2026. We are not affiliated with Anthropic. We wrote this from the documentation and have not tested every habit on every plan.

The free loop-stop skill gives long tasks a limit and a stop rule. If you want a second pair of eyes on your CLAUDE.md and settings, the Tune-Up Kit is a checklist for exactly that.