Spatial Decision-Making on FrozenLake size 8
81.74Success RateGemini-3-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-FlashSetting=Naive2026.06 | 81.74 | |
| SpecFlowSetting=Flow(image) + AR(Text)2026.06 | 81.73 | |
| DiffThinkerSetting=Flow(image) Only2026.06 | 79.83 | |
| GPT-5.1Setting=Naive2026.06 | 79.68 | |
| MVoTSetting=Autoregressive2026.06 | 78.94 | |
| Qwen3-VL-32BSetting=SFT2026.06 | 62.91 | |
| Qwen3-VL-32BSetting=GRPO2026.06 | 59.47 | |
| Qwen3-VL-8BSetting=SFT2026.06 | 53.72 | |
| Qwen3-VL-8BSetting=GRPO2026.06 | 50.21 | |
| Qwen3-VL-32BSetting=Naive2026.06 | 48.12 | |
| Qwen3-VL-8BSetting=Naive2026.06 | 37.25 |