Benchmarks · our own harness · no Elo, no votes

ThreeJSBench

A one-line design prompt → a single-file three.js page. Same pinned three@0.182, 1280×800, headless Chrome.

scoring: side by side, qualified, cost, time · threejs-bench/

Lava Lamp

7 models · 2026-10-06
A lava lamp: glass, wax, and light. Blobs rise, merge and split; lit from the bulb below. Slow and hypnotic, dark room, one hero object.
Opus 5.5 output▶ View live
anthropic logo
Opus 5.5
$0.076m 34s28 fpslive
Sonnet 5.5 output▶ View live
anthropic logo
Sonnet 5.5
$0.011m 55s26 fpslive
GPT 6 Astra output▶ View live
openai logo
GPT 6 Astra
$0.032m 29s7 fpslive
GPT 6 Luna output▶ View live
openai logo
GPT 6 Luna
<$0.0153s2 fpslive
GPT 6 Sol output▶ View live
openai logo
GPT 6 Sol
$0.014m 19s13 fpslive
GPT 6.1 Sol output▶ View live
openai logo
GPT 6.1 Sol
<$0.012m 6s21 fpslive
timeout 20m, no index.html
not qualified
deepseek logo
DeepSeek V4.1 Flash
$0.3820m 0s
Each model runs in its own sandbox through the TODO for AI agent — identical prompt, no view of the others. We publish the raw outputs, what they cost and how long they took. Nothing is voted on, nothing is fitted to a curve. Reproducible from the benchmarks repo.