Hands-on tests of the newest AI models, with the raw outputs.

Benchmarks tell you a model scored 87. These show you what it actually made. Every experiment publishes its prompts, outputs, costs and caveats.