What you will have at the end
You will know which bill your Claude Code use lands on, be able to read the /usage screen, and have a short set of settings that keep context small. Facts come from Anthropic's Claude Code documentation page Manage costs effectively, read on October 31, 2026.
Before you start
Claude Code charges by API token consumption, but who pays depends on how you sign in.
- Pro, Max and Ultra: usage is included in the subscription and limited by plan windows, so the dollar figure in /usage is not a bill.
- Team and Enterprise: each seat has an allowance that resets on a rolling five-hour window and a weekly window, shared with Claude chat and Cowork.
- Claude Console and cloud providers: usage is billed per token to the organization.
- Anthropic's average across enterprise deployments: about $13 per developer per active day, $150 to $250 a month, and under $30 per active day for 90% of users.
1. Read /usage
Run /usage. The Session block shows tokens by model and a total cost computed locally at list price; it resets when /clear starts a new session. Subscribers also see plan usage bars and an attribution breakdown showing the share used by skills, subagents, plugins and each MCP server. Since v2.1.305, when the footer shows a hint such as f to fix, pressing f applies the recommendations it can.
2. Keep context small
Every request sends the conversation so far, so cost grows with context. Use /clear when you switch to unrelated work; it costs nothing, while /compact itself reads the whole conversation it summarises. Move workflow-specific instructions out of CLAUDE.md into skills, which load only when invoked, and keep CLAUDE.md under about 200 lines. Disable unused MCP servers with /mcp and prefer CLI tools such as gh where available.
3. Match model and effort to the task
Anthropic recommends Sonnet for most coding tasks and Opus for complex architecture or multi-step reasoning; switch with /model. Thinking tokens bill as output, so lower effort with /effort for simple tasks. Opus 5.5 and Sonnet 5.5 always use extended thinking, so effort is the lever there. Give simple subagents model: haiku in their configuration. On models with a fixed thinking budget you can also cap it with the MAX_THINKING_TOKENS environment variable; adaptive-reasoning models ignore that setting.
A switch to Opus also applies to subagents that inherit your session's model, so a quick question asked in an Opus session can cost more than you expect once subagents start. Set the model per subagent where the task is routine, such as searching files or summarising logs.
4. Worked example: trim a test-heavy session
Suggested example, not a recorded run. A session that runs the full test suite after each change fills context with logs. Delegate the run to a subagent so only a summary returns, or add a PreToolUse hook that filters test output to failures, as the documentation shows. Vague requests such as improve this codebase trigger broad scanning, while a precise request lets Claude read only the files it needs. Then ask precisely:
5. Check the effect
Run /usage again after a comparable task and compare.
- Fewer input tokens per request and a higher cache share.
- Subagents and MCP servers no longer dominate the attribution list.
- No idle usage from scheduled tasks or goal check-ins you did not need.
- For teams: spend limits set in admin settings, and per-user reports via the analytics export or OpenTelemetry.
Common problems and fixes
Usage climbs in a session left open for hours: long context, cache misses after breaks, scheduled tasks and cross-session messages all resend the full context. Start a fresh session with /clear or turn off the background features you do not use. Agent teams cost more: Anthropic says about seven times the tokens of a standard session when teammates run in plan mode, so keep team tasks small.
The cost in /usage does not match the invoice: figures use list price unless an administrator sets the modelPricing managed setting (v2.1.242 or later). A message such as rate limited until 3:14am in /usage only means plan data cannot be refreshed yet. When you hit a plan limit, /usage-credits lets you buy usage credits or ask an admin.
Plan limits run out sooner than expected: limits depend on model, chat length and file size, according to Claude's pricing page. Check the attribution list in /usage for the largest consumer, then switch routine work to Sonnet, move verbose work to subagents and keep sessions short. Max and Ultra plans offer about five and ten times Pro usage.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.
Go a little further