{"slug": "a-rainbow-slinky-on-an-up-escalator-built-by-15-models", "title": "A rainbow slinky on an up escalator, built by 15 models", "summary": "A benchmark run by Nishaanth Reddy tasked 15 models from Anthropic, OpenAI and Google with a single prompt to build one HTML file showing a rainbow slinky walking down an up escalator in perpetual motion with realistic, unscripted physics and start, stop and reset buttons, with each model given one fresh folder, one prompt and no follow-ups at xhigh reasoning effort (or high where xhigh was unavailable). Build times ranged from 5 minutes for GPT-6.1 Terra at $0.17 in API cost to 301 minutes for Claude Fable 5.1 at $51.50, while Claude Opus 5.5 took 74 minutes at $13.51 and GPT-6.1 Sol took 11 minutes at $0.42. The test followed a Sep 4, 2026 first version whose prompt produced only guided animations, prompting the November prompt to demand realistic, unscripted physics.", "body_md": "# A rainbow slinky on an up escalator, built by 15 models\n\nBy [Nishaanth Reddy](https://morethanamachine.com/authors/nishaanth-reddy/) · \n\nI keep giving new models the same small task. One prompt asks for a single HTML file with a rainbow slinky that walks down an up escalator forever, built with real physics. Fifteen models from Anthropic, OpenAI and Google have taken it so far.\n\n## [00] The prompt\n\nEvery model gets this prompt once, in a fresh folder. Reasoning effort is xhigh, or high where the agent doesn’t offer xhigh.\n\n```\nMake 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.\n```\n\n## setup: agents, effort, recording\n\n```\nSlinky on an up escalator: setup\n\nPrompt (paste once):\nMake 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.\n\nRules\n- One fresh, empty folder per model.\n- Reasoning effort xhigh, or high where the agent doesn't offer xhigh.\n- Send the prompt once. No follow-ups and no edits to the output.\n- Build time runs from sending the prompt to the agent's final answer.\n\nAgents as run\n- Codex CLI 0.167.0 and 0.169.1: model_reasoning_effort = \"xhigh\"\n- Claude Code 2.1.322: /effort xhigh\n- Antigravity CLI 1.5.10: --effort xhigh, or high where xhigh is not offered\n\nRecording\n- Chromium at 1280 x 720 with 2x to 3x pixel density.\n- A fixed 60 Hz virtual clock drives requestAnimationFrame and performance.now.\n- Press Start, let it run 0.2 s, then capture 20 s, cropped to the slinky.\n```\n\n## [01] The runs\n\nClick any two clips below to compare them.\n\n## [02] Time and cost\n\n| Build time and API cost for each of the 15 runs |  |  | \n|---|---|---|\n| Model |  |  | \n|---|---|---|\n| Gemini 4 NeonNov 19 | 25 min | $4.87 | \n| Gemini 4 HydrogenNov 19 | 57 min | $10.51 | \n| Gemini 4 CarbonNov 19 | 58 min | $5.28 | \n| Gemini 4 ArgonNov 19 | 22 min | $2.25 | \n| Gemini 3.8 FlashNov 19 | 21 min | $1.43 | \n| GPT-6.2 AstraNov 17 | 16 min | $2.16 | \n| GPT-6.2 SolNov 17 | 11 min | $0.55 | \n| Claude Opus 5.5Nov 9 | 74 min | $13.51 | \n| Claude Fable 5.2Nov 9 | 64 min | $18.40 | \n| Claude Sonnet 5.5Nov 9 | 97 min | $7.92 | \n| Claude Fable 5.1Nov 9 | 301 min | $51.50 | \n| GPT-6 AstraNov 9 | 14 min | $3.59 | \n| GPT-6.1 SolNov 9 | 11 min | $0.42 | \n| GPT-6.1 TerraNov 9 | 5 min | $0.17 | \n| GPT-6 SolNov 9 | 6 min | $0.44 | \n\nBuilt in is the time from sending the prompt to the agent’s final answer. API cost is what each run’s tokens would cost at that day’s API prices. The runs themselves used subscription plans.\n\n## [0A] Appendix: where it started\n\nThe 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.\n\nLeft: 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.\n\n### The first prompt\n\n```\nMake 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.\n```\n\n", "url": "https://wpnews.pro/news/a-rainbow-slinky-on-an-up-escalator-built-by-15-models", "canonical_source": "https://morethanamachine.com/posts/slinky-escalator-benchmark/", "published_at": "2026-11-20 00:00:00+00:00", "updated_at": "2026-11-21 07:43:20.386572+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-agents", "ai-research"], "entities": ["Nishaanth Reddy", "Anthropic", "OpenAI", "Google", "Claude Opus 5.5", "Claude Fable 5.1", "GPT-6.1 Terra", "Gemini 4 Neon"], "also_reported_by": [], "takeaway": "Across 15 single-prompt runs, build time ranged from 5 minutes (GPT-6.1 Terra, $0.17) to 301 minutes (Claude Fable 5.1, $51.50), and the Sep 4, 2026 prompt without a physics requirement produced only scripted animations.", "recommendation": {"advice": "Run each model once in a fresh folder with a single prompt and no follow-ups, then record the output under a fixed 60 Hz virtual clock to compare build time, API cost and whether the physics is real rather than scripted.", "playbook": "evals-testing", "playbook_url": "https://wpnews.pro/playbooks/evals-testing"}, "alternates": {"html": "https://wpnews.pro/news/a-rainbow-slinky-on-an-up-escalator-built-by-15-models", "markdown": "https://wpnews.pro/news/a-rainbow-slinky-on-an-up-escalator-built-by-15-models.md", "text": "https://wpnews.pro/news/a-rainbow-slinky-on-an-up-escalator-built-by-15-models.txt", "jsonld": "https://wpnews.pro/news/a-rainbow-slinky-on-an-up-escalator-built-by-15-models.jsonld"}}