Spatial Planning on Sokoban
79Success RateAtlasVA
Evaluation Results
| Method | Links | |
|---|---|---|
| AtlasVAModel Category=Ours, Parameter Count=3B2026.05 | 79 | |
| GPT-5Model Category=Proprietary2026.05 | 70 | |
| VAGENModel Category=Open-Source2026.05 | 61 | |
| o3Model Category=Proprietary2026.05 | 60 | |
| Gemini 2.5 ProModel Category=Proprietary2026.05 | 58 | |
| Gemini 2.5 flashModel Category=Proprietary2026.05 | 52 | |
| o4-miniModel Category=Proprietary2026.05 | 44 | |
| GPT-4oModel Category=Proprietary2026.05 | 43 | |
| ActFocusBackbone=3B, Evaluation Step=2002026.05 | 37.5 | |
| ActFocusBackbone=0.5B, Evaluation Step=2002026.05 | 34 | |
| Claude Sonnet 4.5Model Category=Proprietary2026.05 | 31 | |
| Gemini 2.0Model Category=Proprietary2026.05 | 28 | |
| UniformBackbone=3B, Evaluation Step=2002026.05 | 26.2 | |
| Claude Sonnet 3.7Model Category=Proprietary2026.05 | 25 | |
| Qwen2.5-VL-72BModel Category=Open-Source2026.05 | 18 | |
| Qwen2.5-VL-3BModel Category=Open-Source2026.05 | 14 | |
| Qwen2.5-VL-7BModel Category=Open-Source2026.05 | 13 | |
| VLM-R1-3BModel Category=Open-Source2026.05 | 13 | |
| UniformBackbone=0.5B, Evaluation Step=2002026.05 | 12.5 |