Spatial decision-making on Maze size 12
78.31Success RateSpecFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SpecFlowSetting=Flow(image) + AR(Text)2026.06 | 78.31 | |
| Gemini-3-FlashSetting=Naive2026.06 | 76.94 | |
| DiffThinkerSetting=Flow(image) Only2026.06 | 74.86 | |
| MVoTSetting=Autoregressive2026.06 | 72.94 | |
| GPT-5.1Setting=Naive2026.06 | 72.83 | |
| Qwen3-VL-32BSetting=SFT2026.06 | 58.97 | |
| Qwen3-VL-32BSetting=GRPO2026.06 | 55.26 | |
| Qwen3-VL-8BSetting=SFT2026.06 | 48.63 | |
| Qwen3-VL-8BSetting=GRPO2026.06 | 44.92 | |
| Qwen3-VL-32BSetting=Naive2026.06 | 41.92 | |
| Qwen3-VL-8BSetting=Naive2026.06 | 31.39 |