Embodied task planning on BEHAVIOR-1K B-100 1.0
97.5SP%Gemini-3.1-Pro
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini-3.1-ProPrompt style=Guided extended-BDDL2026.06 | 97.5 | 97.5 | 99.19 | 0 | |
| GPT-5.4Prompt style=Guided extended-BDDL2026.06 | 97 | 97 | 99.2 | 1 | |
| D-SFT+DAPO+Short-RLSize=8B, Prompt style=Guided extended-BDDL, Decoding seeds=32026.06 | 95.2 | 95.5 | 99.32 | 1.5 | |
| Gemma-4-31B-ITSize=31B, Prompt style=Guided extended-BDDL2026.06 | 94.5 | 96.5 | 99.21 | 2 | |
| D-SFT+DAPOSize=8B, Prompt style=Guided extended-BDDL, Decoding seeds=32026.06 | 94.3 | 95.7 | 98.67 | 2 | |
| D-SFT+DAPO+Short-RL+GroupAdapt (τ=0.5)Size=8B, Prompt style=Guided extended-BDDL, Decoding seeds=3, τ=0.52026.06 | 94 | 95.2 | 99.22 | 2.2 | |
| Kimi-K2.6Size=1T, Prompt style=Guided extended-BDDL2026.06 | 93 | 97.5 | 99.36 | 4.5 | |
| GLM-5.1Size=754B, Prompt style=Guided extended-BDDL2026.06 | 93 | 96.5 | 98.4 | 5.5 | |
| Qwen3.6-27BSize=27B, Prompt style=Guided extended-BDDL2026.06 | 90.5 | 96.5 | 98.49 | 6.5 | |
| DeepSeek-V4-FlashSize=284B, Prompt style=Guided extended-BDDL2026.06 | 90 | 96 | 98.74 | 6.5 | |
| D-SFTSize=8B, Prompt style=Guided extended-BDDL, Decoding seeds=32026.06 | 87 | 93.5 | 97.7 | 9.5 | |
| DeepSeek-V4-ProSize=1.6T, Prompt style=Guided extended-BDDL2026.06 | 86.5 | 95.5 | 98.24 | 9.5 | |
| Qwen3.6-35B-A3BSize=35B, Prompt style=Guided extended-BDDL2026.06 | 83 | 94 | 96.62 | 15.5 | |
| Qwen3.5-122B-A10BSize=122B, Prompt style=Guided extended-BDDL2026.06 | 82.5 | 93 | 96.93 | 13.5 | |
| Qwen3.5-27BSize=27B, Prompt style=Guided extended-BDDL2026.06 | 80 | 92.5 | 94.91 | 18.5 | |
| Qwen3-8BSize=8B, Prompt style=Guided extended-BDDL2026.06 | 72 | 87 | 95.17 | 23 | |
| Qwen3.5-35B-A3BSize=35B, Prompt style=Guided extended-BDDL2026.06 | 69 | 82.5 | 86.81 | 23.5 |