Documentation
Image studio and model intelligence
Ghost Palette is a creation workspace first, with model comparison and benchmark data built in. These docs explain the workflows, the live ImageBench scores produced on our stack, and public reference data from industry leaderboards.
Live leaderboard
GP-reproduced ImageBench V1 pass rates — latest grade per challenge from community suite runs.
OpenRun suite
Execute the fixed 192-prompt ImageBench suite, VLM-grade pass/fail, and contribute to the live board.
OpenIndustry benchmarks
Curated external rankings — Arena Elo, published ImageBench, GenEval, speed, and price.
Read moreImageBench V1
Published ImageBench methodology, 192-prompt suite structure, and official leaderboard.
Read moreMethodology
Fair comparison rules for Arena and Refine workflows.
Read moreCreate on your prompts
Benchmarks answer fixed suites. The Studio answers your brief — one prompt, model choice when needed, and side-by-side comparison when it helps you decide.
Open Studio