Playground · Animation
The jelly effect: Claude Opus 5.5, GPT‑6.1 Sol, Kimi K3 and Mistral Large 4
A soft jelly that wobbles when you poke it is a small test with a lot in it: a believable physics feel, a good-looking surface, touch and mouse handling, and restraint about what else to put on the page. Each model got the same prompt at its default settings; where one ran out of room we gave it more, and say so. The answers below are exactly what they wrote, running in sandboxed frames; drag them about.
The prompt, word for word
Create the famous jelly effect: a soft, glossy jelly that wobbles and jiggles realistically when you click, tap or drag it, then settles back into shape. Make it a single self-contained HTML file (inline CSS and JavaScript, no external libraries, fonts or images). It should look delightful, fill the viewport, and work with both mouse and touch.
Every answer runs live, in a sandboxed frame. Click into one to play with it.
How each model did
| Model | Our notes | Time | Output tokens | Cost |
|---|---|---|---|---|
| Mistral Large 4 | A glossy jelly dome that pulls into a peak when dragged, splashes droplets, wobbles and is round again within about three seconds: real soft-body physics, and a quiet, tidy page. It thinks a great deal first (64,028 tokens here), so it needed 128,000 tokens of room and 23 minutes. At low thinking effort it finished in 15 minutes, but its jelly flew off-screen and the face slid out of the body.Took 3 tries: ran out of room at 32,000 tokens; ran out of room at 64,000 tokens; finished at 128,000 tokens. | 1386 s | 87,966 64,028 thinking |
$0.184 $0.385 all tries |
| Kimi K3 | A glossy pink ball. Clicking sets its outline rippling nicely, but dragging barely moves or squashes it, so it reads more as a bubble than a jelly. | 268 s | 9,364 4,539 thinking |
$0.141 |
| GPT-6.1 Sol | A jelly dome on a styled page ("Happiness has a little wobble."), with colour swatches. Drag stretches it upwards and it squashes back on release. The fastest and cheapest working answer. | 120 s | 6,388 1,034 thinking |
$0.064 |
| Claude Opus 5.5 Our pick | The most convincing jelly: a moulded jelly on a plate with a face that reacts. It leans a long way when dragged, sags and settles, and the table can be tapped. Colours and sound too. The dearest answer by far: more than ten times GPT's.Took 2 tries: ran out of room at 32,000 tokens; finished at 64,000 tokens. | 362 s | 36,729 22,081 thinking |
$0.735 $1.376 all tries |
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. This is a quick look, not a benchmark: run it again and every answer will differ. For measured results, see our benchmarks.
Verdict
Claude Opus 5.5 made the jelly people mean when they say “jelly effect”, and the only one with character. Mistral Large 4's is the most physical after it, once it was given room to think, but it took 23 minutes. GPT-6.1 Sol came close for a tenth of Opus's price in two minutes, and made the best-designed page around it. Kimi K3's ball looks good but hardly gives.
Two models ran out of room at the playground's old 32,000-token default, mostly on thinking. The playground now gives answers 64,000 tokens by default.
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