Playground · Simulation · HTML · Standing test
Ball bouncing in a spinning hexagon
A popular test from early 2025: a ball bouncing inside a spinning hexagon, with gravity and friction. It looks simple, but the walls are moving, so a correct answer has to account for the wall's speed where the ball hits it. Get that wrong and the ball sinks into a wall, sticks to it or escapes. We watched each answer for 20 seconds and logged the ball and hexagon positions frame by frame. Look at whether the hexagon turns, whether the ball falls and loses a little height with each bounce, and whether it stays inside without sinking into the walls.
The prompt, word for word
Write a program that shows a ball bouncing inside a spinning hexagon. The ball should be affected by gravity and friction, and it must bounce off the rotating walls realistically. Make it a single self-contained HTML file (inline CSS and JavaScript, no external libraries).
Every answer runs live, in a sandboxed frame. Click into one to play with it.
How each model did
| Model | Our notes | Checks | Time | Output tokens | Cost |
|---|---|---|---|---|---|
| GPT-6.1 Sol | A teal hexagon turning slowly (0.55 rad/s), above a bar of sliders and Pause/Reset buttons. The ball has a stripe so you can see it roll, and it bounces lower each time until it rolls along the lowest wall. Correct, but the scene is short and small for the window and the slow spin gives little to watch. | 3 of 3 | 67 s | 4,125 516 thinking |
$0.041 |
| Claude Opus 5.5 Our pick | A violet hexagon spinning at 1 rad/s with a small slider panel. The ball has a spin marker and a short trail. It drops, bounces lower each time and then rolls and hops along whichever wall is lowest. It never left the hexagon or sank into a wall in our 20-second log, and you can grab the ball and throw it. The panel brushes the top-left corner as the hexagon turns. | 3 of 3 | 60 s | 7,358 1,041 thinking |
$0.148 |
| Claude Fable 5.1 | Very close to Opus: a blue hexagon spinning at 1.2 rad/s, a striped ball with a trail and small sparks on impact, sliders, and click-to-drop or drag-to-throw. The bounces are lively and stay inside. The slider panel covers the left corner of the hexagon as it turns past, and this was the most expensive answer. | 3 of 3 | 84 s | 7,772 2,228 thinking |
$0.390 |
| Gemini 3.8 Flash | The physics works: the ball falls, bounces off the turning walls and never escapes. A drawing bug, though, means the bright glowing stroke is drawn over six spokes from the centre rather than around the outline. So you see a spinning asterisk the ball seems to pass through, while the real walls are only a faint edge of the dark fill.
|
2.5 of 3 | 67 s | 8,612 3,430 thinking |
$0.032 |
| Muse Spark 1.3 | A thick, glowing cyan hexagon turning at 0.6 rad/s, with a pink ball, a trail and a slider panel. Bounces are correct and fade to rolling along the bottom walls. The panel sits over the left part of the hexagon, so the ball is sometimes hidden behind it. | 3 of 3 | 90 s | 6,300 2,290 thinking |
$0.027 |
| Grok 4.7 | A thin violet hexagon spinning anticlockwise at 0.9 rad/s. The wall the ball touches lights up amber, and a live read-out shows speed and position. The ball spends most of its time bouncing and rolling in the lower-right corner, where the spin carries it, and it never leaves. Drag-to-throw works, but the panel covers the hexagon's left side. | 3 of 3 | 215 s | 13,691 9,017 thinking |
$0.083 |
| Kimi K3 | The plainest answer: a cyan hexagon turning at about 0.6 rad/s, an orange ball and a "click anywhere to move the ball" hint, with no controls. A few bounces quickly die down, and for most of the 20 seconds the ball rolls back and forth along the bottom walls. Correct, but quiet. | 3 of 3 | 152 s | 16,763 12,259 thinking |
$0.218 |
| Qwen3.8 Max (0902) | The most elaborate page, "HEX CHAMBER", with a lab-style dashboard counting velocity, ball spin and impacts, plus sliders and Pause, Reset and Kick. The 1.2 rad/s spin, the bounces and the rolling are all correct. The controls panel covers a large part of the hexagon's left side and the title overlaps the top, and the answer took over 12 minutes. | 3 of 3 | 727 s | 29,663 19,938 thinking |
$0.178 |
| DeepSeek V4.1 Flash | A pale-outlined hexagon spinning at 0.85 rad/s, with an orange ball and a curved trail, and no controls. The ball bounces, then rolls with enough speed to ride up the left wall to above the centre before dropping off, which is how a real ball would behave. Correct physics for under a cent. | 3 of 3 | 87 s | 16,815 13,620 thinking |
$0.009 |
| Mistral Large 4 | "HEXA BOUNCE": a cyan hexagon spinning fast (2.2 rad/s), with a red ball that has a spin highlight and a trail, a speed read-out and an option to make gravity turn with the hexagon. The fast walls keep throwing the ball back up, so it stays lively rather than settling, and it never got through a wall. It took over 18 minutes to answer. | 3 of 3 | 1118 s | 50,196 36,605 thinking |
$0.105 |
One answer per model, at each model's default settings unless the card says otherwise, through OpenRouter. If an answer ran out of room or stopped short, we asked again with more room, then with thinking off, and say so. We tried every answer, with Claude Opus 5.5 running the checks and drafting the notes; it is an Anthropic model, so we check its calls on Claude answers ourselves. This is a quick look, not a benchmark: run it again and every answer will differ. For measured results, see our benchmarks.
What we checked, by trying each answer ourselves: The hexagon rotates; The ball falls under gravity and bounces, losing a little energy; The ball stays inside and bounces off the moving walls (no tunnelling through or getting stuck). A partial pass counts as half.
Verdict
Every model produced a hexagon that turns and a ball that falls, bounces and loses height, and in our 20-second logs no ball escaped or got stuck in a wall. The differences are in presentation. Claude Opus 5.5 is our pick: correct moving-wall bounces, a ball you can see roll, a panel that mostly stays clear of the hexagon, and drag-to-throw, in 60 seconds for $0.148. Claude Fable 5.1 is nearly identical but cost $0.390. Qwen3.8 Max has the richest page, but its controls cover part of the hexagon and it took 727 seconds.
For value, DeepSeek V4.1 Flash gave a correct simulation with no controls for $0.0095, and GPT-6.1 Sol a tidy one for $0.041. Gemini 3.8 Flash is the one visible failure: its physics is right, but a drawing bug turns the bright outline into six spokes, so the real walls are hard to see. Mistral Large 4 also works but took 1,118 seconds.
What would your prompt get?The Spring Prompt playground runs one prompt on up to six models from OpenAI, Anthropic, Google, Mistral and others, side by side.
Ask for access
More from the playground
