THE FINE PRINT DEPARTMENT
ABOUT THE INDEX
TokenBurn Index is the public index of AI token waste: a measuring instrument wearing an institutional suit. The suit is a joke. The measurements are not.
What this is
AI bills are quiet. Nothing tells you that turn 40 of a chat re-pays for turns 1–39, that the same prompt costs 70x more on one model than another, or that your agent's “thinking” bills by the token. This site makes the meter visible: a calculator that counts and judges your prompts, a monthly audit for your usage CSV, an index of expensive habits, and a Hall of Shame for the incidents nobody expenses proudly.
How the numbers work
- — Token counts use the o200k_base encoder in your browser: exact for GPT models, ~±10% for other vendors, labeled as such.
- — Prices for the 12 tracked models are checked daily against public list prices by an automated watcher (last change: 2026-07-11). Changes above sanity thresholds are reviewed by a human before publishing.
- — Privacy: prompts and CSVs are processed entirely client-side. No upload, no storage, no login. We don't want your invoices.
- — The Burn Score and the grades are editorial. The formula is documented, deliberately opinionated, and applied consistently. The arithmetic underneath it is not editorial.
- — Estimates are labeled as estimates, savings are “up to”, and no environmental salvation is promised at any point.
Who runs this
Llorenç Ballester
Builds applied AI, automation and software for a living — and started counting other people's tokens after watching his own agents think out loud at his expense. Every score, verdict and judgment on this site is his fault. The arithmetic, however, checks out.
Contact
Corrections, pricing errata and Hall of Shame confessions: [email protected] · X: @tokenburnindex. Corrections are applied; confessions are anonymized.