Fine But Why
Newsletter

Index

Benchmarks

One prompt. Then I open the file, or the render, and write down what is actually there.

  1. Muse Wonder Four games in /goal mode A lawn, a patio, a space station and snakes in the grass. One prompt each: keep refining until it looks beautiful.
  2. Opus 5.5 and Wonder The kitchen test Same prompt, two tools. Then two players, skills, and a player who drops out mid-cook.
  3. Opus 5.5 xhigh A house from one prompt 2 h 15 min, 1,206 lines, zero asset files. Every texture is code.
  4. Grok 4.7 xhigh A table in Blender Whole terminal on the left, whole Blender window on the right.
  5. Grok 4.7 and Opus 5 The second try saved the file The browser stayed open. Grok low, 362 seconds. Opus high, 468. The sun landed first.