Spatial decision-making on MiniBehavior size 20
72.96Success RateSpecFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SpecFlowSetting=Flow(image) + AR(Text)2026.06 | 72.96 | |
| Gemini-3-FlashSetting=Naive2026.06 | 70.19 | |
| DiffThinkerSetting=Flow(image) Only2026.06 | 66.58 | |
| GPT-5.1Setting=Naive2026.06 | 66.14 | |
| MVoTSetting=Autoregressive2026.06 | 65.23 | |
| Qwen3-VL-32BSetting=SFT2026.06 | 48.72 | |
| Qwen3-VL-32BSetting=GRPO2026.06 | 44.35 | |
| Qwen3-VL-8BSetting=SFT2026.06 | 38.94 | |
| Qwen3-VL-8BSetting=GRPO2026.06 | 35.02 | |
| Qwen3-VL-32BSetting=Naive2026.06 | 31.63 | |
| Qwen3-VL-8BSetting=Naive2026.06 | 23.96 |