THE TICKER · CHECKED DAILY · LAST CHANGE 2026-09-06
AI API PRICING
What 25 frontier models actually charge, in USD per million tokens. The same request costs 63x more on GPT-5.6 Cyber than on GPT-5.6 Luna. Most teams never look. You are looking.
METHODOLOGY: effective public price — promos included — cross-checked daily against the LiteLLM community dataset with per-model pinned keys and sanity guards (a claimed 3x+ overnight change gets human review, not autopilot). Your negotiated rates may differ. Cache, batch and reasoning rates are the provider's published numbers, not a rule of thumb — where a provider publishes none, we say so instead of assuming one. Full methodology → · Price changes →
THE BOARD
EVERY HEAD-TO-HEAD →“Output multiple” is what a token costs leaving the model vs. entering it — the reason verbose answers hurt more than verbose questions. The last column prices a 1,000-token prompt with a 3,000-token reply (editorial assumption, same as the calculator's prompt profile).
PRICE CHANGE LOG
Claude Opus 5 joins the board at $5.00 in / $25.00 out per MTok — priced identically to Opus 4.8.
The ticker goes live. Prices checked daily against public list prices from here on.
Every applied change lands here automatically the day the daily check catches it. No entry, no change — silence is also data. Machine-readable: RSS feed.
USE THE DATA
The whole board plus the changelog, as JSON, free, no key, no rate limit worth mentioning. Updated on every price change. If you use it somewhere public, link back — that's the entire license.
curl -s https://tokenburnindex.com/pricing.json
Schema: { updated, unit, criteria, models[], changelog[] } — prices in USD per 1M tokens.
Or embed a live price badge in your README — it updates itself when vendors move:
[](https://tokenburnindex.com/pricing/claude-sonnet-5)
Works for any model id on the board: /badge/pricing/<id>.svg
PRICES ARE HALF THE STORY
The other half is how many tokens you burn to get an answer. Paste a prompt into the Burn Calculator to see what it costs on every model above — or check what your habits rank on the AI Waste Habits.