Aquarium: Haiku 5.5 high vs GPT-6.1 Sol medium
0.0 severy clip in sync · same script, same moment
The numbers
What the robot tested
The promptthe same for every model · one shot · no tools
Every model on this test → How we test →
The clips: our robot opens each model's page and plays the same script (same clicks at the same instant, same clock, same random seed), recorded frame by frame at 30 fps. So they say nothing about smoothness: "fps live" is what the robot measured in a real browser. The numbers belong to the run in the clip. Cost is API-equivalent.
What is the robot test?
Passing means the page works as the prompt asked. It does not judge looks: that one is your call.