Reports

Head-to-head tests, with the data.

Each report compares real systems on the same inputs and publishes the numbers, the method, and what we would do with the result.

• Judge duel · September 23, 2026

  Which small model can catch a three-armed dinosaur?

  We tested open typed-decision models and a frontier API as judges for deformed AI images, on public data and on a live book service. Accuracy, false alarms, speed, and what fits on one 24 GB GPU.