ccusage vs tokscale vs whoburnedmore

The three tools share a common origin: they all parse the usage logs that AI coding assistants write to your local disk, then convert raw token counts into an estimated dollar cost. Where they diverge is in what happens after that parse — whether the numbers are kept local, submitted to a leaderboard, ranked by tokens or cost, and whether you can build a social identity around your usage.

This comparison is written to help you pick the right tool for your workflow, not to declare a winner. All three are genuinely useful; they differ in scope and philosophy.

How each tool works

ccusage (ryoppippi/ccusage)

ccusage is the open-source library and CLI that started the category. Run npx ccusagein any directory and it scans your machine for Claude Code, Codex, Gemini CLI, and a growing list of other AI tool log directories. It prints a per-tool, per-day token and cost table directly in your terminal. Nothing is ever sent anywhere. It's pure local analysis — deliberately so. The library is also importable as an npm package, so you can build your own scripts on top of it. ccusage is maintained actively, well-documented, and has the broadest tool coverage of the three.

tokscale (junhoyeo/tokscale)

tokscale takes ccusage-style local parsing and pairs it with a public leaderboard at tokscale.ai. It supports multiple AI coding tools, submits your totals to a shared ranking, and lets you build a public profile. The leaderboard ranks by cost — whoever has spent the most money in aggregate sits at the top. tokscale is the most direct competitor to whoburnedmore in terms of the social feature set, though the two tools differ in ranking dimension, privacy controls, and the social graph model.

whoburnedmore

whoburnedmore uses ccusage as its data layer (the parser is literally the ccusage npm package), then adds a public leaderboard, developer profiles linkable to an X handle, a friends-board social graph, and a --local flag for fully private mode. The ranking is by tokens burnedrather than cost — a unit that doesn't advantage developers on cheaper models. Profiles are claimable without an account: run one command with your X handle and the entry is yours.

Full feature comparison

The table below covers the features developers ask about most when choosing between the three tools. “Yes” means the feature is available and works today (June 2026).

How each tool ranks developers

The ranking dimension matters more than it looks. Ranking by cost (tokscale) means developers using expensive models like Claude Opus rise faster than developers on the same absolute token volume but cheaper models. Ranking by tokens(whoburnedmore) treats a million Claude tokens the same as a million Codex tokens — the score is pure volume.

Neither is wrong. Cost-ranking answers “who spends the most on AI coding?” Token-ranking answers “who burned the most raw compute?” If you're trying to win a leaderboard, the dimension determines which model mix wins.

Privacy model differences

This is where the tools differ most philosophically. ccusage never sends data anywhere; the binary keeps all analysis on your disk. tokscale submits your cost totals to its leaderboard; there is no documented local-only mode. whoburnedmore submits token and tool totals to the leaderboard by default, but ships a first-class --local flag that skips submission entirely — the browser dashboard opens from a local server and nothing leaves your machine.

On the default (submission) path, whoburnedmore sends per-tool token counts and timestamps. It never sends prompt text, file names, or code snippets. You can run npx whoburnedmore --dry-run to see the exact JSON payload before committing.

Which tool should you use?

Choose ccusage if…

You want zero network egress, you need a scriptable npm library, you care most about detailed per-day and per-model breakdowns, or you work in an environment where outbound HTTP is restricted. ccusage is the safest choice when privacy is non-negotiable and you don't need a leaderboard. 🛡️

Choose tokscale if…

You want a public leaderboard ranked by dollar spend and a quick profile without needing a friends-board or X-handle integration. If cost is your primary metric and you're happy with a submission-by-default model, tokscale is a solid pick.

Choose whoburnedmore if…

You want to rank by raw tokens burned, link your leaderboard profile to your X handle, build a friends board with colleagues, or need the flexibility to run privately with --localwhen on client machines. whoburnedmore is also the right choice if you already trust ccusage's parser — the numbers are produced by the same code.

For a closer look at whoburnedmore as a ccusage replacement for the specific use-case of adding a leaderboard, see the ccusage alternative guide. To understand how the leaderboard itself works and what it costs to join the top ranks, read the AI coding token leaderboard guide.