How to check Claude Code usage: tokens, cost and plan limits

Run /usage inside Claude Code. It shows this session's tokens and estimated cost, your plan's usage bars, and what has been eating them. That answers "how much right now", not "how much this month, on which project".

Check usage inside Claude Code

/usage is the one command; /cost and /stats open the same screen. What it shows depends on how you pay:

• Session block — tokens per model, cache reads and writes, and a dollar figure Claude Code computes locally at list price. Anthropic's docs say it is "intended for API users": on Pro and Max the dollar figure is not what you are billed. It resets when /clear starts a new session.
• Plan usage bars (subscription plans) — how much of your five-hour and weekly windows is used.
• Breakdown (same plans) — what counted against the limits: skills, subagents, plugins, MCP servers, scheduled loops, and behaviours such as long context or cache misses. d and w switch between 24 hours and 7 days.

The breakdown is "computed from local session history on this machine", so another laptop or claude.ai is not in it. /status is not a usage command — it shows version, model and account.

See token usage per session and per model

The session block lists tokens per model for the session you are in. For older sessions there is no built-in token or cost view: /usage covers the current session and a 24-hour or 7-day window, not a session from last Tuesday. /insights reviews recent sessions, but it reports how you work, not what they used. And the transcripts all of this is read from are deleted after 30 days by default.

Which number is your bill

• On an API key, the Claude Console's usage page is the authoritative bill, in tokens and spend; its Claude Code dashboard shows spend and accepted lines per member, and the Claude Code Analytics API returns the same per day. An API key has no plan, so there are no plan windows.
• On Team and Enterprise, owners get a spend report in org analytics: spend per user and per model for usage credits, updated daily. Usage inside the seat allowance is not metered in dollars, and nothing is per project.
• On Bedrock, Google Cloud or Microsoft Foundry, your cloud's billing console is the bill.

How much of my Claude Code usage limit is left?

The bars in /usage show how much of the five-hour and weekly windows is used and when each resets, Settings > Usage on claude.ai shows the same for the account, and the status line gets rate_limits.five_hour and rate_limits.seven_day on Pro and Max.

The /usage bars are the official answer. They are your account's windows: Claude Code, claude.ai and Claude Desktop draw from the same allowance. To watch them without opening anything, put them on the status line: Claude Code passes the five-hour and seven-day windows to it on Pro and Max, next to how full the context is and the session's cost at list price. The line is on screen for as long as the session runs and keeps nothing after it. The status line guide shows how. What each plan allows, and what it costs, is in Claude Code pricing. How the windows work and reset is in Claude usage limits.

Usage monitors and trackers compared

Three tools read the same local transcripts and keep going where /usage stops. The table at the top of this page lines them up; read 7 October 2026:

• ccusage runs through npx or bun x and leaves no background process. It prints daily, weekly, monthly, per-session and five-hour-block reports in tokens and dollars, for Claude Code and most other coding agents, and --instances groups Claude Code by project. Its five-hour blocks are worked out from your transcripts, not read from Claude Code.
• Claude-Code-Usage-Monitor is a live terminal view of the current five-hour window — tokens, cost, burn rate and a forecast of when you will run out — for as long as its terminal is open. It has daily and monthly views, and an opt-in local warehouse that keeps history by project and model past the 30-day cleanup. It uses the status line's plan figures when they are fresh and its own estimate otherwise.
• phuryn/claude-usage is a local web dashboard of token use, cost and session history for Claude Code.

Five monitors, each with what it is best for and its limits, are compared in Claude Code usage monitors.

Every session's usage, in a desktop app

Caprock is a free desktop app for macOS, Windows and Linux that records every Claude Code session on the machine, whether or not it started it — and Codex, OpenCode and DeepSeek Harness next to them, plus the Gemini CLI sessions it starts; not Cursor or Copilot. It keeps the record past the 30-day cleanup, and every dollar it shows is a list-price estimate, as /usage is, beside the plan's dollars and the share of the weekly limit the work used. Its Cost screen answers what /usage cannot:

• Any range — today, 7 days, 30 days or everything recorded.
• Per project — cost and tokens for each repository, and per directory inside a monorepo.
• Per model — every model's turns, tokens and cost, Claude and GPT side by side, priced at list price.
• What the money went on — running commands, writing code, reading and searching, MCP tools, and turns with no tool call, with a note when tool calls could not be matched to a turn.
• Plan value — on a flat plan, what the same work would cost at API list price, with a caveat beside it that this is not a bill and not money saved.

The plan windows are on the same screen once caprock statusline install is set up. caprock report prints the same totals as text you can paste into an issue or a post, with the caveat attached.

What it cannot show: an absolute token ceiling for your plan, because Claude Code does not report one — only the percentage used. And like /usage, it sees this machine only.