UniG2U-Bench (Generate-then-Answer) - Puzzles and Games: leaderboard
Metric: Accuracy (%) on the 537 puzzle and game items (mental reconstruction and tracking, visual tracking, maze, jigsaw and sliding puzzles), from UniG2U-Bench, Generate-then-Answer: the unified model first generates an intermediate image (an auxiliary drawing or state) and then answers from the original and generated images, greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Bagel | 31.58 | |
| 2 | Uni-Video | 29.33 | |
| 3 | Show-o2 | 26.39 | |
| 4 | UniWorld-V1 | 24.72 | |
| 5 | UniPic2 | 23.95 | |
| 6 | STAR-7B | 23.56 | |
| 7 | OmniGen2 | 21.91 | |
| 8 | OneCAT-3B | 20.69 | |
| 9 | ILLUME+ | 19.53 | |
| 10 | Ovis-U1-3B | 18 |
Interactive version: theaggregate.ai/benchmark?slug=unig2u-bench-generate-then-answer-puzzles-and-games · How It Works · Data refreshed daily, snapshot 2026-10-11.