Spatial decision-making on Maze size 16
64.85Success RateSpecFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SpecFlowSetting=Flow(image) + AR(Text)2026.06 | 64.85 | |
| Gemini-3-FlashSetting=Naive2026.06 | 61.27 | |
| DiffThinkerSetting=Flow(image) Only2026.06 | 59.42 | |
| GPT-5.1Setting=Naive2026.06 | 58.91 | |
| MVoTSetting=Autoregressive2026.06 | 57.81 | |
| Qwen3-VL-32BSetting=SFT2026.06 | 46.35 | |
| Qwen3-VL-32BSetting=GRPO2026.06 | 42.17 | |
| Qwen3-VL-8BSetting=SFT2026.06 | 36.85 | |
| Qwen3-VL-8BSetting=GRPO2026.06 | 33.27 | |
| Qwen3-VL-32BSetting=Naive2026.06 | 29.84 | |
| Qwen3-VL-8BSetting=Naive2026.06 | 21.07 |