Same prompt, many models. You judge by eye.

Each page is one visual or spatial prompt, run by different models at different reasoning-effort levels. Open one to judge their work side by side, with cost, time, tokens and the full agent transcript.