The AI Coding Token Leaderboard: Who Burned Most?

Token burn is the honest metric for AI coding intensity. It doesn't depend on which subscription tier you picked or how expensive your model is — it measures raw compute volume. The developer at the top of the board burned the most tokens, full stop. whoburnedmore makes that number visible, comparable, and shareable.

How does the leaderboard ranking work?

Every entry on the board is a real developer who ran npx whoburnedmore and submitted their local log totals. The ranking is purely by cumulative token count — the sum of all tokens across every AI coding tool whoburnedmore detected on that machine. Input tokens and output tokens are both counted; cached tokens are noted separately but included in the total.

Rankings update every time you submit — typically each time you run the command. The board is live and public at whoburnedmore.com; anyone can browse it without logging in. 🔥

Why tokens, not dollars?

A cost-based ranking would privilege developers who use expensive models. If you burn 100M tokens on Claude Opus and I burn 100M tokens on Gemini Flash, we worked equally hard — but my bill is a fraction of yours. Token-ranking puts us in the same bracket. It's a fairer measure of AI coding intensity. If you prefer a cost-ranked leaderboard, check out tokscale — both serve valid use cases.

How do I get on the leaderboard?

You don't need an account. One command is all it takes:

1. 1

   Run npx whoburnedmore in your terminal

   The CLI scans your machine for AI coding tool log directories (Claude Code, Codex, Gemini CLI, Cursor, opencode, Amp, Goose, and more) and aggregates the token counts by tool and by day.
2. 2

   Review the local summary

   You see a breakdown in the terminal before anything is submitted. Token totals, estimated cost at API rates, and which tools contributed most.
3. 3

   Totals are submitted to the board

   Only the aggregate numbers go to the server: per-tool token count, timestamps, and (optionally) your X handle if you pass --claim @yourhandle. No code, prompts, or file names are ever sent.
4. 4

   Your entry appears on the leaderboard

   Open whoburnedmore.com and find your entry. First-time submissions start at the bottom and climb as more of your history is included.

$ npx whoburnedmore↳ scanning usage logs on this machine…   TOOL            TOKENS      EST. COST  claude code     91.2M       $303.40  codex           38.7M        $92.20  gemini cli      22.4M        $18.50  cursor           9.1M        $31.00  ──────────────────────────────────────  total          161.4M      $445.10   → submitting totals to leaderboard…  → rank: #212 of 3,104 developers 🏆  → profile: whoburnedmore.com/d/anon-8f3c

How do I claim my leaderboard profile?

By default, your entry appears under a random anonymous slug like /d/anon-8f3c. To pin your real identity to it, pass your X username on first or any subsequent run:

$ npx whoburnedmore --claim @yourhandle↳ linking @yourhandle via X OAuth…  profile now at: whoburnedmore.com/d/yourhandle

Claiming is optional. Your entry is on the board either way. Claiming just replaces the random slug with your handle so friends can find you, and allows you to be followed in friends boards.

Friends boards

Once you have a claimed profile, other developers can add your handle to their friends board — a private mini-leaderboard that shows only the people they follow. It's a great way to run a team burn-rate competition or keep tabs on your most prolific AI-coding colleagues. To see how that fits into the broader tool comparison, read ccusage vs tokscale vs whoburnedmore.

Can I participate without being public?

Yes. The --local flag skips submission entirely and opens your full dashboard in a local browser window. You get the same per-tool breakdown and charts — just no leaderboard entry and no data leaving your machine. This is the right mode for client machines, shared workstations, or any time you want the numbers without the social layer.

What does it take to crack the top 100?

The threshold shifts as the board grows, but in mid-2026 the top 100 developers have each burned somewhere north of 150 million tokens across all tools. That sounds like a lot — and it is — but a single intensive week of Claude Code on a large codebase can burn 20–30M tokens on its own. Developers running multiple agents simultaneously climb the board fast.

The per-tool breakdown also matters for understanding your own score. Most high-ranking developers run Claude Code as their primary agent (it's the highest-volume tool in most setups) with Codex or Gemini CLI as a secondary. To read more about what drives token volume per tool, see how to check your AI coding token usage.

Ready to see where you land? One command puts you on the board. Use npx whoburnedmore --local first if you want to preview your numbers in private before publishing.