Spatial decision-making on MiniBehavior size 12
91.27Success RateSpecFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SpecFlowSetting=Flow(image) + AR(Text)2026.06 | 91.27 | |
| Gemini-3-FlashSetting=Naive2026.06 | 90.36 | |
| DiffThinkerSetting=Flow(image) Only2026.06 | 88.31 | |
| GPT-5.1Setting=Naive2026.06 | 87.45 | |
| MVoTSetting=Autoregressive2026.06 | 86.78 | |
| Qwen3-VL-32BSetting=SFT2026.06 | 72.16 | |
| Qwen3-VL-32BSetting=GRPO2026.06 | 69.04 | |
| Qwen3-VL-8BSetting=SFT2026.06 | 63.18 | |
| Qwen3-VL-8BSetting=GRPO2026.06 | 59.41 | |
| Qwen3-VL-32BSetting=Naive2026.06 | 56.71 | |
| Qwen3-VL-8BSetting=Naive2026.06 | 46.28 |