Sokoban on Sokoban
55Success RateGemini 3.1 Pro
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 3.1 ProGroup=Closed (zero-shot)2026.05 | 55 | 21 | |
| Adaptive (Qwen2.5-VL-7B)Group=Ours2026.05 | 35.9 | 30 | |
| GPT-5.5Group=Closed (zero-shot)2026.05 | 35 | 22.1 | |
| Claude SonnetGroup=Closed (zero-shot)2026.05 | 0 | — | |
| InternVL3-8B/14B/78BGroup=Open (zero-shot)2026.05 | 0 | — | |
| Qwen2.5-VL-7B/72BGroup=Open (zero-shot)2026.05 | 0 | — | |
| Qwen3-VL-8B/32BGroup=Open (zero-shot)2026.05 | 0 | — |