Which AI Has the Best Physics Feel? (Gyro Phone Test)

Four “Thinking” models. One identical animation prompt. The difference in smoothness is brutal. This 2x2 showdown compares Cursor Composer 1 vs Claude Opus 4.5 (Thinking) vs ChatGPT 5.2 (Thinking) vs Gemini 3 Pro (Thinking). They all attempt the same single-file HTML/CSS/JS Canvas build: a floating smartphone that tilts and subtly yaws, with a neon gyroscope ring and UI elements that lag/overshoot using spring-damper inertia. Watch for the tells: perfectly seamless looping, stable frame rate, believable easing, clean glow blending, and crisp phone edge highlights. Comment which quadrant you think wins before the end, and drop the next prompt you want tested. Prompt: "Create a single-file HTML/CSS/JS animation of a floating smartphone that continuously tilts left/right and subtly yaws, while a neon “gyroscope” ring and UI elements inside react with inertia (slight lag/overshoot) as if the phone has real mass. Use only Canvas 2D (no libraries), simulate a simple spring-damper for inertia, add soft glow (additive blending), and ensure a seamless 6–8 second loop with a dark gradient background and crisp highlights on the phone edges." Music provided by pixabay