Spatial decision-making on MiniBehavior size 16
79.84Success RateSpecFlow
Evaluation Results
| Method | Links | |
|---|---|---|
| SpecFlowSetting=Flow(image) + AR(Text)2026.06 | 79.84 | |
| Gemini-3-FlashSetting=Naive2026.06 | 78.02 | |
| DiffThinkerSetting=Flow(image) Only2026.06 | 75.21 | |
| GPT-5.1Setting=Naive2026.06 | 73.96 | |
| MVoTSetting=Autoregressive2026.06 | 73.88 | |
| Qwen3-VL-32BSetting=SFT2026.06 | 60.41 | |
| Qwen3-VL-32BSetting=GRPO2026.06 | 56.78 | |
| Qwen3-VL-8BSetting=SFT2026.06 | 50.27 | |
| Qwen3-VL-8BSetting=GRPO2026.06 | 46.38 | |
| Qwen3-VL-32BSetting=Naive2026.06 | 44.28 | |
| Qwen3-VL-8BSetting=Naive2026.06 | 34.41 |