Embodied Reasoning on ALFWorld Seen
96.5Pick & Place Success RateRESKILL
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RESKILLModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 96.5 | 81.4 | 94.9 | 90.2 | |
| SkillRLModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 95.5 | 82.8 | 87.2 | 89 | |
| GRPOModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 93.8 | 62.3 | 53.8 | 74.8 | |
| GRPOModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 93.8 | 72.1 | 66.7 | 80.7 | |
| RESKILLModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 91.6 | 88.7 | 76.9 | 90 | |
| INSPOModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 90.4 | 79.2 | 71.8 | 83.2 | |
| Sonnet 4.5Model Scale=Sonnet 4.5, Strategy=Proprietary2026.06 | 89.8 | 73.5 | 96.2 | 82.5 | |
| MemRL (+RL)Model Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 89.8 | 75.8 | 71.8 | 81.2 | |
| SkillRLModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 89.8 | 85.9 | 61.5 | 85.7 | |
| EvolveRModel Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 88.7 | 79.8 | 74.4 | 83.3 | |
| MemRL (+RL)Model Scale=Qwen3-8B, Strategy=w/ RL2026.06 | 86.4 | 78.5 | 84.6 | 83.1 | |
| Best‡Model Scale=Qwen3-8B, Strategy=w/ Evolve2026.06 | 79.1 | 58.3 | 61.5 | 66.2 | |
| Best†Model Scale=Qwen3-8B, Strategy=w/ Skill2026.06 | 79.1 | 56.4 | 64.1 | 66.4 | |
| INSPOModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 78 | 77.6 | 61.5 | 76.4 | |
| ReActModel Scale=Qwen3-8B, Strategy=Baseline2026.06 | 73.4 | 44.8 | 64.3 | 59.3 | |
| EvolveRModel Scale=Qwen3-4B-Instruct, Strategy=w/ RL2026.06 | 70.6 | 85.2 | 61.5 | 76.7 | |
| Best†Model Scale=Qwen3-4B-Instruct, Strategy=w/ Skill2026.06 | 61.6 | 43.1 | 61.5 | 51 | |
| Best‡Model Scale=Qwen3-4B-Instruct, Strategy=w/ Evolve2026.06 | 60.5 | 44.6 | 79.5 | 54.5 | |
| ReActModel Scale=Qwen3-4B-Instruct, Strategy=Baseline2026.06 | 54.7 | 28.1 | 28.6 | 37.1 |