How many tokens does Claude Code actually burn?

Claude Code is the most effective way ever invented to convert electricity into pull requests, and also into tokens. A single agentic session can read dozens of files, run tools, retry failures and spawn subagents — every step a model call, every call carrying context. Most people have no idea what that adds up to. The fun part: your machine has been keeping receipts the whole time.

The local tally

Claude Code maintains aggregate usage stats on your machine, per model, in ~/.claude/stats-cache.json— total input and output tokens, among other counters. No cloud dashboard required; it's a JSON file you can open right now. If the numbers in there look implausible, remember how agent context compounds: the agent re-sends its working set on every call, so a day of heavy use in a large repo racks up tokens at a rate no chat window ever could.

What is that worth in money?

If you're on a Claude subscription, those tokens are covered by a flat price — which is exactly why the raw count is so entertaining. Priced at API list rates (what the same tokens would cost pay-as-you-go, e.g. $15/M output for Sonnet-class models, $25–50/M for the frontier tiers), a serious month of Claude Code often "costs" more than the subscription by an order of magnitude. That notional value is the number worth flexing: what you would have burned.

Read yours in one command — no account needed

Burnmaxxing ships a zero-install importer that does the arithmetic for you — it reads the local stats file, sums tokens per model, applies list prices, and uploads only the aggregate:

curl -fsSL https://www.burnmaxxing.com/import.mjs | node -

No signup, no API key, no login. It prints two links: a public card showing your burn, and a private claim link. The card is shareable immediately; the claim link is what turns it into an account and a leaderboard rank, whenever you feel like it.

• Everything is computed locally. What leaves your machine is two numbers: total tokens and their notional USD value. No logs, no prompts, no code, no file contents.
• Already signed in? Pass the import token from your profile and it goes straight onto your card, skipping the claim step.
• Re-running it replaces your previous Claude Code snapshot, so you can update whenever the number gets more embarrassing.

(Piping a URL into nodeshould always make you pause — the script is served from this site, and it's short enough to read first. Please do.)

Then what?

Your Claude Code burn lands next to your API-verified spend on the global leaderboard, anonymous by default, and flows into your Burnmaxxer Card. Somewhere out there is a developer whose agent burned a billion tokens last month. Find out how far behind them you are — or aren't. If it's "aren't", congratulations: you're absolutely burnmaxxing.