Tangram puzzle solving (Rotation) on TANGRAMPUZZLE (Unseen)
43.2Strict GCQwen-Image-Edit-20B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen-Image-Edit-20BInput=Text + Image, Output=Image2026.01 | 43.2 | 85.7 | 99.6 | |
| Wan2.2-TI2V-5BInput=Text + Image, Output=Video2026.01 | 22.4 | 74.5 | 98 | |
| Nano BananaInput=Text + Image, Output=Image2026.01 | 9.8 | 43.4 | 64.7 | |
| Qwen3-VL-8BInput=Text + Image, Output=Image2026.01 | 1.6 | 52.1 | 80.8 |