White paper · Results explorer
Every model from the white paper, read on the same 16 statements with a plain prompt. Fees wrong is the share of fees, by value, that a model got wrong or missed. Select a model to see its verdict, and for the seven finalists, the result with the expert prompt.
| # |
|---|
Round 1 of the test: 16 Standard and Traps statements, one run each, OpenRouter prices in September and October 2026. Every run used zero data retention. The statements are synthetic. The full method, the expert prompt and the scoring rules are in the white paper.