Leaderboard
Bradley-Terry ratings from 8,214 pairwise human votes. Ties split the point; “both bad” carries no signal.
| # | Competitor | Rating | W | L | T | Games |
|---|---|---|---|---|---|---|
| 1 | GPT-5.5 (naive prompt) — benchmark | 1116 | 2423 | 871 | 93 | 3387 |
| 2 | Prompt v0 · baseline | 951 | 1241 | 2059 | 104 | 3404 |
| 3 | Prompt v2 · accuracy+clean | 933 | 961 | 1695 | 103 | 2759 |
By category
Characters & Mascots
GPT-5.5 (naive prompt) — benchmark
Rating 1206 · 651 games
Data Visualization
GPT-5.5 (naive prompt) — benchmark
Rating 1093 · 458 games
Technical Diagrams
GPT-5.5 (naive prompt) — benchmark
Rating 1124 · 642 games
Icons & Glyphs
Prompt v2 · accuracy+clean
Rating 1021 · 609 games
Logos & Wordmarks
GPT-5.5 (naive prompt) — benchmark
Rating 1198 · 387 games
Single Objects
GPT-5.5 (naive prompt) — benchmark
Rating 1156 · 591 games
Patterns & Textures
Prompt v2 · accuracy+clean
Rating 1067 · 358 games
Scenes & Illustrations
GPT-5.5 (naive prompt) — benchmark
Rating 1194 · 634 games
UI Mockups
GPT-5.5 (naive prompt) — benchmark
Rating 1143 · 445 games