Newshot

A Slinky on an Escalator Is a Weirdly Good Test of LLM Intelligence

Original text(via HN AI)
题图
题图

I started this as a follow-up to the pelican-on-a-bicycle test, pushed toward inter-object physics. Every time a new model drops, I ask it for a single HTML file of a rainbow slinky walking down an up escalator forever, with real physics instead of a scripted animation. Fifteen models from Anthropic, OpenAI and Google have tried it so far.

It’s a silly test, though it’s turned into a surprisingly good gut-check for how capable these models actually are. Getting a slinky to flip down moving steps without stalling or flying off takes spring physics, step collisions and enough self-testing to spot when the result looks off.

[00] The Prompt

Each model gets this exact prompt once in an empty folder, with no follow-ups or manual edits. I set reasoning effort to xhigh, or high when the agent doesn’t offer xhigh.


        Make a single HTML file of a rainbow slinky walking down an up escalator, in perpetual motion. Make it realistic and not a scripted animation. Have a start, stop and reset button.
      

        Slinky on an up escalator: setup

Prompt (paste once):
Make a single HTML file of a rainbow slinky walking down an up escalator, in perpetual motion. Make it realistic and not a scripted animation. Have a start, stop and reset button.

Rules
- One fresh, empty folder per model.
- Reasoning effort xhigh, or high where the agent doesn't offer xhigh.
- Send the prompt once. No follow-ups and no edits to the output.
- Build time runs from sending the prompt to the agent's final answer.

Agents as run
- Codex CLI 0.167.0 and 0.169.1: model_reasoning_effort = "xhigh"
- Claude Code 2.1.322: /effort xhigh
- Antigravity CLI 1.5.10: --effort xhigh, or high where xhigh is not offered

Recording
- Chromium at 1280 x 720 with 2x to 3x pixel density.
- A fixed 60 Hz virtual clock drives requestAnimationFrame and performance.now.
- Press Start, let it run 0.2 s, then capture 20 s, cropped to the slinky.
      

[01] The Runs

图表:xhigh · 1 h 14 min · $13.51

图表:xhigh · 25 min · $4.87

Click any two clips below to compare them.

Even with “not a scripted animation” in the prompt, several models still bend the rules to keep the slinky going. A few adjust the escalator speed to hold it in frame. Others add a small torque assist or reset the slinky if it stalls.

These are single runs across three different agents (Codex CLI, Claude Code and Antigravity CLI), recorded on a virtual 60 Hz clock so heavy sims don’t drop frames. It’s purely an anecdotal check.

[02] Time And Cost

Gemini 4 Neon Nov 1925 min$4.87
Gemini 4 Hydrogen Nov 1957 min$10.51
Gemini 4 Carbon Nov 1958 min$5.28
Gemini 4 Argon Nov 1922 min$2.25
Gemini 3.8 Flash Nov 1921 min$1.43
GPT-6.2 Astra Nov 1716 min$2.16
GPT-6.2 Sol Nov 1711 min$0.55
Claude Opus 5.5 Nov 974 min$13.51
Claude Fable 5.2 Nov 964 min$18.40
Claude Sonnet 5.5 Nov 997 min$7.92
Claude Fable 5.1 Nov 9301 min$51.50
GPT-6 Astra Nov 914 min$3.59
GPT-6.1 Sol Nov 911 min$0.42
GPT-6.1 Terra Nov 95 min$0.17
GPT-6 Sol Nov 96 min$0.44

Build time runs from sending the prompt to the agent’s final answer, including any self-testing the agent did along the way. API cost is what each run’s tokens would have cost at that day’s API prices, though I ran all of these on subscription plans.

[0A] Appendix: First Test

The first version of this test ran on Sep 4, 2026 with a different prompt. Both pages that came out of this first prompt were guided animations, so the November prompt asks for realistic, unscripted physics instead.

Left: GPT-6 Astra, max effort, Codex CLI. Right: Claude Fable 5.1, Claude Code. The video labels both sides "max". The Claude session most likely ran at xhigh.

First Prompt


        Make me a single HTML file of a rainbow slinky going down an up-escalator forever. No libraries — just canvas and code. The slinky should be a chain of springs, each coil a different color of the rainbow. It starts folded in half like a horseshoe, draped over a step. When dropped it flips end-over-end down the steps, and because the escalator keeps moving up, it tumbles in place and never reaches the bottom. Include a drop button and a reset button.
      

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by HN AI
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗