Sequential Planning on Visual Spatial Planning (VSP) FrozenLake
100VSP FrozenLake Success Rate (Level 3)Gemini-3-Flash
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Gemini-3-FlashSetting=N/A2025.12 | 100 | 100 | 100 | 99 | 83 | 98 | |
| DiffThinker++Setting=Flow Matching2025.12 | 100 | 100 | 100 | 98 | 100 | 100 | |
| GPT-5Setting=N/A2025.12 | 99 | 70 | 67 | 43 | 36 | 29 | |
| Qwen3-VL-8BSetting=SFT2025.12 | 99 | 96 | 98 | 96 | 92 | 86 | |
| Qwen3-VL-32BSetting=GRPO2025.12 | 99 | 90 | 95 | 69 | 73 | 58 | |
| DiffThinkerSetting=Flow Matching2025.12 | 99 | 100 | 98 | 99 | 100 | 100 | |
| Qwen3-VL-32BSetting=SFT2025.12 | 96 | 99 | 98 | 100 | 99 | 90 | |
| Qwen3-VL-8BSetting=GRPO2025.12 | 91 | 70 | 70 | 31 | 34 | 24 | |
| Qwen3-VL-32BSetting=N/A2025.12 | 75 | 51 | 47 | 25 | 23 | 26 | |
| Qwen3-VL-8BSetting=N/A2025.12 | 64 | 46 | 33 | 21 | 12 | 21 | |
| Qwen-Image-Edit-2511Setting=N/A2025.12 | 50 | 55 | 44 | 16 | 23 | 23 | |
| Qwen-Image-Edit-2509Setting=N/A2025.12 | 33 | 36 | 22 | 12 | 11 | 7 |