About the joke with a database
The benchmark no one asked for and everyone understands instantly.
BuseyBench tests creative model behavior with a deliberately specific prompt: generate a recognizable Gary Busey portrait. It is not a general intelligence benchmark. It is a public record of prompt following, visual memory, tool policy, artifact validity, and model weirdness over time.
Why this exists
Most benchmark charts are abstract. This one is visual, funny, repeatable, and easy to inspect. If a model says it can generate SVG, BuseyBench asks it to prove it in the most facially demanding way available.
What it measures
The main signal is not just likeness. It also measures whether the model follows constraints, produces valid files, avoids remote assets, handles teeth and expression, and improves as model releases change.
Independence note
BuseyBench is independent, editorial, and not endorsed by Gary Busey or any model provider. Generated images are benchmark artifacts, not real photos.