FAQs
We crawled 2.7 million websites, found 36,000 real pricing pages, then asked
43 different AI models to recreate them. The result is 1,000,000 pricing cards
you can scroll through from $1/mo to $1,000,000/mo. Every button is real and charges real money.
What is this?
An infinite scroll of AI-generated pricing pages. We scraped real pricing pages from
across the internet, extracted their designs, colors, and copy, then fed them to 43 AI
models (GPT-5, Claude, Gemini, Llama, and 39 others) as inspiration. Each model
generated thousands of unique pricing sections. They're stacked cheapest to most expensive.
Wait, the buttons actually work?
Yes. Click any "Get Started," "Subscribe," or "Buy Now" button on any card and it opens a
real Stripe checkout for that card's exact price. You are buying nothing. There is no product.
Your payment shows up on the public leaderboard.
Why does this exist?
Every website has a pricing page now. Open source projects, side projects, AI wrappers,
landing pages for things that don't exist yet. The entire internet is a shopping mall.
This is what happens when you take that to its logical extreme: one million pricing pages,
generated by AI, where every button actually charges you.
Which AI made the best pages?
Claude Opus and GPT-5 made the most polished designs. Llama 3.3 70B consistently produced
the worst. GPT-3.5 Turbo was surprisingly decent given it's the cheapest model.
Liquid LFM 1.2B (1.2 billion parameters) made chaotic pages with giant fonts and
white-on-white text. It cost $600 total across all 43 models.
Why is there a leaderboard?
Because paying real money for public rank on a joke website is inherently funny.
The leaderboard shows who paid the most. Higher payment = higher rank. That's it.
How was this built?
Crawled 2.7M URLs with Crawlee. Classified 903K pages with Gemini Flash Lite ($55).
Extracted 22K design blueprints with Playwright. Generated 111K HTML fragments across
OpenAI Batch, Anthropic Batch, Doubleword, and OpenRouter. Validated with parse5.
Frontend is vanilla TypeScript with a custom virtual scroller rendering 339K rows
in Shadow DOM. Data is served from Cloudflare R2. The whole thing took 2 weeks.