Spatial decision-making on Maze size 4
94.38Success Rate (%)Gemini-3-Flash
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3-FlashSetting=Naive2026.06 | 94.38 | |
| SpecFlowSetting=Flow(image) + AR(Text)2026.06 | 94.12 | |
| DiffThinkerSetting=Flow(image) Only2026.06 | 93.1 | |
| MVoTSetting=Autoregressive2026.06 | 92.03 | |
| GPT-5.1Setting=Naive2026.06 | 91.76 | |
| Qwen3-VL-32BSetting=SFT2026.06 | 86.21 | |
| Qwen3-VL-32BSetting=GRPO2026.06 | 83.02 | |
| Qwen3-VL-8BSetting=SFT2026.06 | 78.92 | |
| Qwen3-VL-8BSetting=GRPO2026.06 | 75.31 | |
| Qwen3-VL-32BSetting=Naive2026.06 | 72.18 | |
| Qwen3-VL-8BSetting=Naive2026.06 | 61.72 |