Maze Navigation on MAZENAVIGATION 7x7 Long OOD Both 1.0
4,000EMWan2.2-TI2V-5B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Wan2.2-TI2V-5BInput=Text + Image, Output=Video, Evaluation Protocol=Fine-tuned2026.01 | 4,000 | 5,110 | |
| Wan2.2-TI2V-5B (Unseen Visual Icons)Input=Text + Image, Output=Video, Evaluation Protocol=Fine-tuned2026.01 | 3,800 | 4,790 | |
| GPT-5.1Input=Text + Image, Output=Text, Evaluation Protocol=Zero-shot2026.01 | 0 | 0 | |
| GPT-5.2Input=Text + Image, Output=Text, Evaluation Protocol=Zero-shot2026.01 | 0 | 0 | |
| Qwen3-VL-8BInput=Text + Image, Output=Text, Evaluation Protocol=Fine-tuned2026.01 | 0 | 1,130 | |
| Qwen3-VL-8B (w/ coordinates)Input=Text + Image, Output=Text, Evaluation Protocol=Fine-tuned2026.01 | 0 | 810 | |
| VPRL-7B *Input=Image, Output=Image, Evaluation Protocol=Fine-tuned2026.01 | 0 | 410 |