Skill Selection on SkillsBench 1000-skill scale
36.8Reward (%)SkillDAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SkillDAGBackbone Model=gpt-5.2-codex2026.06 | 36.8 | 75.9 | 70.1 | 73 | |
| GoSBackbone Model=gpt-5.2-codex2026.06 | 34.4 | 65.5 | 50.6 | 57.3 | |
| Vanilla SkillsBackbone Model=gpt-5.2-codex2026.06 | 27.4 | — | — | — | |
| SkillDAGBackbone Model=MiniMax-M2.72026.06 | 27.3 | 78.2 | 66.7 | 71.3 | |
| Vector SkillsBackbone Model=gpt-5.2-codex2026.06 | 21.5 | 10.8 | 3.6 | 5.8 | |
| GoSBackbone Model=MiniMax-M2.72026.06 | 18.7 | 65.5 | 50.6 | 57.3 | |
| Vanilla SkillsBackbone Model=MiniMax-M2.72026.06 | 17.2 | — | — | — | |
| Vector SkillsBackbone Model=MiniMax-M2.72026.06 | 10.4 | 10.8 | 3.6 | 5.8 |