Interactive Agent Task Completion on SocialMaze
82.7Average RewardOpenSkill
Evaluation Results
| Method | Links | |
|---|---|---|
| OpenSkillTarget Agent=Opus 4.62026.06 | 82.7 | |
| No SkillTarget Agent=Opus 4.62026.06 | 81.6 | |
| Skill CreatorTarget Agent=Opus 4.62026.06 | 81 | |
| Self-GenTarget Agent=Opus 4.62026.06 | 80.4 | |
| MementoTarget Agent=Opus 4.62026.06 | 80.3 | |
| CoTTarget Agent=Opus 4.62026.06 | 79.4 | |
| AutoSkillTarget Agent=Opus 4.62026.06 | 77.2 | |
| OpenSkillTarget Agent=GPT 5.22026.06 | 70.7 | |
| Skill CreatorTarget Agent=GPT 5.22026.06 | 69.8 | |
| Self-GenTarget Agent=GPT 5.22026.06 | 67.6 | |
| MementoTarget Agent=GPT 5.22026.06 | 67.4 | |
| CoTTarget Agent=GPT 5.22026.06 | 66.7 | |
| No SkillTarget Agent=GPT 5.22026.06 | 66.6 | |
| AutoSkillTarget Agent=GPT 5.22026.06 | 65.9 |