Today's prices
Same tasks for every model. Each price is its real invoice divided by the work it finished.
bar length & colour = price
quality = work finished
▲▼rank vs price per token
Quality is the share of work a model finished. Prices are real OpenRouter invoices, every token and retry counted.
How we appraise
Every price comes out of the same machine — a task worth a fixed amount of work, solved on a real invoice, graded pass or fail, then divided into one honest number.
- Task Worth a fixed number of work units. Fix a failing test5 WU
- Model Solves it; we keep the real invoice. spent $0.0021
- Grader Scores it 0 to 1, mechanically. passed · 5/5 WU
- Price Invoice ÷ work earned. $0.42 / 1k WU
Failure still costs. Miss the grader and the invoice stays on the bill with zero work earned — so the price climbs. A model that fails cheaply isn't cheap.