Claude Code costs: checking usage and cutting spend

Who this is for

People who want to know roughly what Claude Code costs, or who need to find out why a bill jumped. That covers individual subscribers, API users and team admins. After reading, you can read the current session's usage in /usage, find where team spend is reported, and adjust the settings that matter most.

How billing works

Claude Code charges by API token consumption. Subscription plans (Pro, Max, Ultra, Team and Enterprise) are priced at claude.com/pricing. Per-developer costs vary widely with the model you choose, the size of the codebase, and habits such as running several instances or automation.^[1]

The official docs give one reference point for enterprise deployments: an average of about $13 per developer per active day, or $150 to $250 per developer per month, with 90% of users below $30 per active day. Treat it as an average, not a forecast for your own use.^[1]

Reading the session with /usage

Run /usage and the Session block at the top shows token usage for the current session. That block is intended for API users. Pro, Max and Ultra subscribers have usage included in their plan, so the session cost has no direct billing meaning for them.^[1] Output looks like this:^[1]

Total cost:            $0.55
Total duration (API):  6m 20s
Total duration (wall): 6h 33m 10s
Total code changes:    0 lines added, 0 lines removed
Usage by model:
   claude-sonnet-4-6:  1.2k input, 5.3k output, 940.0k cache read, 50.0k cache write ($0.55)

The dollar figure is computed locally from token counts at list price, unless an organization has set a modelPricing table, in which case the contracted rates apply. It is an estimate. Running /clear starts a new session and resets these totals. The status line's cost field shows the same total.^[1]

After the first API response in the main conversation, /usage adds a Prompt cache (main) line. It reports the request count, the share of input tokens served from cache, cache misses, and whether the cache is still warm. That line requires Claude Code v2.1.251 or later.^[1]

On Pro, Max, Ultra, Team and Enterprise plans, /usage also attributes recent usage to skills, subagents, plugins and individual MCP servers, each shown as a percentage. Press d or w to switch between the last 24 hours and the last 7 days.^[1] These figures are approximate and come from session history on this machine, so usage from other devices or claude.ai is not counted.^[1]

How teams see spend

• Claude for Teams and Enterprise: each member's usage draws from a per-seat allowance that resets on a rolling five-hour window and on a weekly window. Admins see a spend report in org analytics, broken down by user and model, updated daily and exportable as CSV.^[1]
• Claude Console (API): spend is managed through workspaces. The first time you sign in to Claude Code with a Claude Console account, a workspace named "Claude Code" is created automatically.^[1]
• Amazon Bedrock, Google Cloud or Microsoft Foundry: usage is billed per token to your cloud account, and budgets and limits are set in the cloud provider's billing console.^[1]
• OpenTelemetry: works on every setup, and it is the only option that streams per-user token and cost metrics into your own observability stack in near real time.^[1]

If your organization pays contracted rates, the managed setting modelPricing makes Claude Code report figures at those rates. It requires Claude Code v2.1.242 or later. It changes what Claude Code reports, not what Anthropic charges.^[1]

When a subscriber hits the limit

Usage credits let you keep working past your plan's usage limit. Manage them with /usage-credits after you sign in with your claude.ai subscription through /login. The command is not available with API key authentication.^[1]

Ways to cut usage

• Delegate verbose work to subagents. Running tests, fetching documentation or processing logs in a subagent keeps the long output out of the main conversation. The subagent's own requests still count toward your usage.^[1]
• Match the model to the job. Sonnet handles most coding tasks well and costs less than Opus. Reserve Opus for complex architectural decisions and multi-step reasoning. Switch mid-session with /model, and for simple subagent tasks specify model: haiku in the subagent configuration.^[1]
• Keep CLAUDE.md short. CLAUDE.md is loaded into context at session start. Skills load only when invoked, so moving detailed workflow instructions into skills keeps the base context smaller. The docs suggest keeping CLAUDE.md under 200 lines.^[1]
• Lower thinking when it is not needed. Extended thinking is on by default, and thinking tokens are billed as output tokens. For simpler tasks, lower the effort level with /effort, or disable thinking in /config.^[1] On models with a fixed thinking budget, MAX_THINKING_TOKENS=8000 lowers that budget. Adaptive-reasoning models ignore nonzero budgets, so use effort levels there instead.^[1]
• Mind MCP overhead. MCP tool definitions are deferred by default, so only tool names and server instructions enter the context until Claude uses a specific tool.^[1]

Agent teams

Agent teams are disabled by default. Set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 to enable them. When teammates run in plan mode, token usage is about seven times that of a standard session, so keep team tasks small and self-contained.^[1]

Next steps

For the exact billing rules and team settings, use the official cost management documentation.^[1]