Sequential Embodied Task Performance on ALFWorld 140-episode
97.9Average Reward (%)GoS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GoSModel=Claude Sonnet 4.52026.04 | 97.9 | 27,215 | 49.2 | |
| Vector SkillsModel=Claude Sonnet 4.52026.04 | 93.6 | 28,407 | 37.8 | |
| GoSModel=GPT-5.2 Codex2026.04 | 93.6 | 46,462 | 64.7 | |
| Vector SkillsModel=GPT-5.2 Codex2026.04 | 92.9 | 34,436 | 57 | |
| Vanilla SkillsModel=Claude Sonnet 4.52026.04 | 89.3 | 1,524,401 | 53.2 | |
| Vanilla SkillsModel=GPT-5.2 Codex2026.04 | 89.3 | 1,435,614 | 83.3 | |
| GoSModel=MiniMax M2.72026.04 | 54.3 | 65,227 | 68.8 | |
| Vector SkillsModel=MiniMax M2.72026.04 | 50.7 | 66,109 | 73.4 | |
| Vanilla SkillsModel=MiniMax M2.72026.04 | 47.1 | 2,184,823 | 88.6 |