Playground · Simulation · HTML · Standing test
Newton's cradle
A Newton's cradle looks simple but is hard to fake: when you pull out the end ball and let go, the ball at the far end should leap out while the three in the middle stay almost still, and each swing should be a little lower than the last until the row rests. We dragged the leftmost ball out with the mouse in every answer and watched for three things: does the ball follow the mouse on its string, does the momentum pass cleanly through the middle balls, and does the motion die down gradually rather than stopping dead or running for ever.
The prompt, word for word
Make a Newton's cradle with five balls. Drag a ball (or several) out to the side and let go; the swing should pass through the line of balls realistically, losing a little energy each time until it comes to rest. Make it a single self-contained HTML file (inline CSS and JavaScript, no external libraries, fonts or images).
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 | Gold balls on a slim steel frame, with an "In motion" status light and an energy bar that falls as the cradle loses energy. Dragging and release work and the far ball swings out, but the inner balls do not hold still: the second and fourth balls drift with the end balls, opening visible gaps in the row. The fastest answer, at 95 seconds and $0.059.
|
2.5 of 3 | 95 s | 5,929 1,034 thinking |
$0.059 |
| Claude Opus 5.5 Our pick | A wide chrome cradle on a wooden base, with buttons to lift one, two or three balls, a "1 + 1" mode, a restitution slider (0.985) and a sound toggle. The dragged ball follows the mouse on its string, the far ball leaps out while the middle three stay put, and each swing is a little lower until the row is still by 20 seconds. | 3 of 3 | 148 s | 15,414 7,686 thinking |
$0.309 |
| Claude Fable 5.1 | A chrome frame drawn with a second, rear top bar for depth and V-shaped double strings; it starts with the end ball already swinging. Drag and transfer work cleanly, and the swing dies down over about eight seconds. The most expensive answer, at $0.97. | 3 of 3 | 222 s | 19,293 12,726 thinking |
$0.966 |
| Gemini 3.8 Flash | A small cradle on a wooden base with buttons for one, two or three balls, a "Clash (1 vs 1)" mode and sound. The transfer is clean, but the frame is narrower than the swing, so the end balls pass straight through the uprights. Damping is very light: the end balls are still swinging after 40 seconds. | 3 of 3 | 118 s | 15,372 8,949 thinking |
$0.058 |
| Muse Spark 1.3 | A plain grey cradle in a card, with a reset button and a "click" sound option. It behaves correctly: the far ball swings out, the middle balls stay still, and the swing fades slowly, still visible at 20 seconds and at rest by 40. | 3 of 3 | 197 s | 14,304 10,037 thinking |
$0.061 |
| Grok 4.7 | A warm wooden frame with double strings on each ball, on a brown background. The swing passes through cleanly and the motion has nearly died away at 20 seconds. Its first try sent no code after 712 seconds, so it needed a second, which took another 520 seconds.Took 2 tries: sent no code at 64,000 tokens; finished at 64,000 tokens. | 3 of 3 | 520 s | 38,176 32,405 thinking |
$0.230 $0.444 all tries |
| Kimi K3 | Large, softly lit balls in a tall dark frame with a reset button. The transfer is clean and the swing fades slowly over about 40 seconds, though the end balls swing out past the frame's uprights. | 3 of 3 | 267 s | 17,683 11,193 thinking |
$0.248 |
| Qwen3.8 Max (0902) | The most elaborate page: a slanted, close-up frame with motion blur, a "Principia 1687" title card, counters for elapsed time, impacts, peak swing and energy, and lift buttons. The physics holds up once its opening demo settles, but the title card covers the top bar, and the swing loses most of its energy within about ten seconds. The slowest answer, at 729 seconds. | 3 of 3 | 729 s | 28,148 18,049 thinking |
$0.169 |
| DeepSeek V4.1 Flash | A neat cradle with V-shaped double strings inside a wide dark frame. Dragging, a clean transfer and a gradual fade to rest in about seven seconds all work. The cheapest answer, at $0.027. | 3 of 3 | 277 s | 47,158 41,976 thinking |
$0.027 |
| Mistral Large 4 | Large chrome balls hanging from a thin bar on a plain dark page. The physics is sound (clean transfer and a slow fade to rest), but a layout bug places the row so its last ball sits at the centre of the page, leaving the leftmost ball half off the left edge, so it swings out of view when you pull it.
|
2.5 of 3 | 497 s | 21,684 16,745 thinking |
$0.045 |
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: You can pull a ball out and let it go; The far ball swings out while the middle ones stay put; The swinging slowly dies down. A partial pass counts as half.
Verdict
Most models got the physics right. Claude Opus 5.5 is our pick: a clean transfer through still middle balls, a gentle fade to rest, and useful extras (lift buttons and a restitution slider), for $0.31 in 148 seconds. DeepSeek V4.1 Flash, Grok 4.7, Muse Spark 1.3 and Kimi K3 behave just as well with fewer controls; DeepSeek is the cheapest at $0.027, though Grok needed two tries totalling over 20 minutes. Qwen3.8 Max built the most elaborate page, but its title card covers the frame and it took 729 seconds.
GPT-6.1 Sol was the fastest (95 seconds, $0.059) and looks polished, but its inner balls drift apart instead of staying put. Gemini 3.8 Flash's balls swing through the frame's uprights and are still moving after 40 seconds, and Mistral Large 4's cradle is drawn off-centre, so the leftmost ball sits half off the screen.
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
