See Token Usage in Claude Code: /usage, /stats and How Many Tokens You Have Left

TL;DR /usage tells you how much of your plan is left, where it went, and when it resets. /stats and /cost are not other commands: they are aliases of that one and open the same screen. If what you want is the breakdown of the session you have open right now, that one is genuinely different: /context.

First, because it orders everything else: /usage, /cost and /stats are the same command. Claude Code's own code declares it without ambiguity:

{ name: "usage", aliases: ["cost", "stats"],
  description: "Show session cost, plan usage, and activity stats" }

Whichever one you type, the same screen opens. There is no /stats dashboard separate from the /usage gauge. There is one screen with several sections.

What it actually tells you

You can read it without opening the interface, and that way you get both windows at once:

$ claude -p "/usage"

You are currently using your subscription to power your Claude Code usage

Current session: 14% used · resets Aug 16 at 9:20pm (Europe/Madrid)
Current week (all models): 29% used · resets Aug 20 at 8pm (Europe/Madrid)
Current week (Fable): 29% used · resets Aug 20 at 8pm (Europe/Madrid)

What's contributing to your limits usage?

Last 24h · 317 requests · 6 sessions
  95% of your usage was at >150k context
  Top skills: /artifact-diagramming 8%, /claude-code-tip 7%, /linkedin-daily 7%
  Top MCP servers: wmedia 18%

Last 7d · 1566 requests · 15 sessions
  87% of your usage was at >150k context
  59% of your usage came from sessions active for 8+ hours
  37% of your usage came from subagent-heavy sessions
  Top MCP servers: wmedia 9%, plugin:chrome-devtools-mcp:chrome-devtools 7%

That is my machine, untouched. Three things worth reading properly:

The three bars are different windows. The session one is a rolling five hours. The weekly one is what locks you out for days. And the third is specific to the model you use, so switching with /model does not hand your session back: how the two limits actually work.

The attribution block names the culprit: your skills, subagents, plugins and MCP servers, each with a percentage. It is the part of the screen you can act on directly: which skill and which MCP are eating your limit.

The dollar figure is for API users. On Pro or Max your usage is included in the subscription and that number bills nothing. Claude Code computes it locally at list rates, so it reflects no discounts and will not match your real invoice.

Does looking at it cost tokens?

This tip used to get this wrong, and here is the correction: it said no, and the right answer is a little yes. The official docs file it under background consumption:

Command processing: Some commands like /usage may generate requests to check status

We are talking about pennies, typically under $0.04 per session counting everything Claude Code does in the background. That is no reason to stop looking. But "zero" was false, and in a tip about measuring consumption that word matters.

The one that is genuinely free is /clear. The one that is very much not is /compact, because summarizing the conversation means reading all of it.

How to use it

1. Both windows at once, without the interface

claude -p "/usage"

It prints the 24 hours and the 7 days back to back, with the request and session count for each. The interactive screen always opens on the last 24 hours and you have to press w for the week.

2. The interactive screen

> /usage

Or /stats, or /cost. Inside, d and w switch between day and week, and r retries when the limits request fails (which happens when the usage endpoint is rate limited: you will see Showing last-known usage with data up to an hour old).

3. Move it to your status line

To avoid running it every time, put it in your status line and it sits in front of you without typing anything.

4. And /context for the session you are in

> /context

This one really is a different command, answering a different question. /usage is your plan; /context is this conversation: how much the system prompt, tools, MCP servers and messages take up inside the window of whichever model you have selected.

Reference

When to use which

Official docs: Commands reference | Manage costs effectively

Related tips: Monitor context with /context | 10 habits to save tokens | Choose the right model | Prompt caching | The Agent SDK credit that was announced and cancelled | How your usage limits actually work (5h + weekly) | Your limit always in the status line