Interactive Agent Task Completion on ScienceWorld
90Average RewardOpenSkill
Evaluation Results
| Method | Links | |
|---|---|---|
| OpenSkillTarget Agent=Opus 4.62026.06 | 90 | |
| SkillNetTarget Agent=Opus 4.62026.06 | 88.7 | |
| Skill CreatorTarget Agent=Opus 4.62026.06 | 88.3 | |
| Self-GenTarget Agent=Opus 4.62026.06 | 88 | |
| CoTTarget Agent=Opus 4.62026.06 | 88 | |
| No SkillTarget Agent=Opus 4.62026.06 | 87.1 | |
| OpenSkillTarget Agent=GPT 5.22026.06 | 85.3 | |
| Skill CreatorTarget Agent=GPT 5.22026.06 | 83.1 | |
| CoTTarget Agent=GPT 5.22026.06 | 80 | |
| No SkillTarget Agent=GPT 5.22026.06 | 78 | |
| SkillNetTarget Agent=GPT 5.22026.06 | 77.8 | |
| Self-GenTarget Agent=GPT 5.22026.06 | 75.9 |