About
hype or code shows what new AI models make. Models on OpenRouter released in the last 21 days that cost $25 or less per million output tokens get the same six prompts, one shot each. Free routes always qualify. The files go up as delivered, with time, tokens and cost.
Outputs, not scores
Benchmarks are useful. They are also numbers a lab can tune for. A watch that ticks, a Breakout you can play or a pelican on a bicycle is easy to judge with your own eyes. So the six daily prompts get no score and no ranking. You look and decide.
The pelican
The pelican is Simon Willison's test: "Generate an SVG of a pelican riding a bicycle." The other five prompts follow the same idea with more ordinary jobs: a watch, a landing page, a game, a logo and 150 words of copy.
Everyday prompts
Most people ask these models about a letter, an email or dinner, so every text model also gets nine Everyday prompts and every image model four, typed the way people type them. Each holds one thing from real life a model can get wrong, such as a book that doesn't exist. Every answer is shown as it would land on a phone, and you can pick the one you'd use in a blind vote on each prompt's page.
Contact
On X: @analyzestripes. A missing model, a broken run, a prompt you would add: reply or DM there.