How to check your Claude usage: claude.ai, Claude Code and the API
Short answer: it depends on how you pay. On a Claude plan, look at Settings > Usage on
claude.ai, or run /usage in Claude Code. On the API, open the Usage and Cost pages
in the Claude Console.
None of these is a per-conversation token counter. Below: what each screen shows, what it leaves out, and how to get a record of every response if you need one.
On claude.ai and the desktop app (Pro, Max, Ultra)
Open Settings > Usage. Anthropic's help page says you can see "how much you've used and the time remaining under Current session" there, along with your next weekly reset time.
- Two limits apply. A session limit that resets every five hours, and weekly limits that reset once a week. Some plans also have a separate weekly limit for one model.
- One allowance, every surface. Usage on claude.ai, Claude Code and Claude Desktop counts toward the same limit. A heavy Claude Code afternoon leaves less for chat.
- No token count per chat. Anthropic's help pages describe how much of your limit is used, not the number of tokens in the conversation in front of you.
What moves the bar, according to the same page: the length and complexity of your conversations, the features you use (tools and connectors are token-heavy), which model you chat with, and the effort level. Long chats cost more per message because the model rereads the whole conversation on every turn, so starting a new chat once the old context no longer matters is the simplest way to use less.
In Claude Code
Run /usage at the Claude Code prompt. What you see depends on how you signed in:
- On a plan: usage bars for your plan limits, activity stats, and a breakdown of what counted against
the limits. The breakdown attributes recent usage to skills, subagents, plugins and MCP servers, and suggests
changes for anything behind 10% or more of it. Press
dorwto switch between the last 24 hours and the last 7 days. - On an API key: the Session block at the top shows the current session's tokens by model (input, output, cache read, cache write) and a dollar figure.
Two limits on those numbers, both stated in Claude Code's documentation:
- The dollar figure is computed on your machine from token counts at list price. The docs call it an estimate and point to the Console for authoritative billing. On a plan it isn't relevant to billing at all.
- The breakdown is approximate and comes from session history on this machine only. Usage from other devices or from claude.ai isn't in it.
Other places to look while you work:
/contextshows what is taking up the context window right now. A full context window is a different thing from a usage limit.- The count beside the spinner, as in
(10m 27s · ↓ 5.5k tokens), is the output of the current turn only. It leaves out everything Claude Code sends to the model, which is usually the larger part. - The status line can show usage and cost continuously if you configure it.
On the API (Claude Console)
API usage is pay-as-you-go, and the Console is the authoritative record:
- Usage page (
platform.claude.com/usage): input and output tokens by model, date and API key, with filters for workspace, model, month or day, and key. You can click a bar to drill down to hours and minutes, and download the view as CSV. - Cost page (
platform.claude.com/cost): daily cost per model and workspace, with totals that include web search and code execution, and a CSV export.
Both are visible to the Developer, Billing and Admin roles. Anthropic's article notes one gap: these pages can't break usage or cost down by individual user.
On a Team or Enterprise plan
Admins get a spend report in the organization's analytics with estimated spend per user and per model, exportable as CSV. The report covers usage credits; usage inside the seat allowance isn't metered in dollars. Each member's allowance resets on a five-hour window and a weekly window, shared with Claude chat.
What none of these screens give you
- A record per response. The plan screens show a share of a limit. The Console shows totals by model and key. Neither lists each request with its tokens and cost.
- Repeats. The same request sent twice is two requests in every total. Nothing marks the second one. See how to spot duplicate AI API charges.
- One view across companies. Each provider reports only its own usage.
For Claude Code there is a local source you can read yourself: the logs in ~/.claude/projects
record the usage block of every response. Adding those up directly over-counts, because the same response is
written more than once. Why Claude Code token counts come out wrong
explains the fix, and the free counter does it in your browser without uploading
anything.
Free one-page checklist: AI API bill checklist (PDF), ten checks to print or share.
A record of every request, kept for you
NeverPayTwice is a small app we make. It reads your Claude Code logs and counts each response once, and its live meter keeps a receipt for every API request you route through it (Anthropic, OpenAI, Google Gemini and compatible APIs): time, model, request ID, tokens and cost. It flags requests you paid for twice. It never blocks or changes a request, nothing is uploaded, and API keys are never stored. Mac, Windows or Linux, $9 one time.
See it with sample data first, no sign-up.