SkillsBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
SkillsBench (test)
48.9Reward
30
SkillsBench
56.9Overall Success Rate (avg@5)
26
SkillsBench 11 domains
44.8Overall Score
16
SkillsBench 86 runnable tasks (frozen-policy transfer)
79Valid Task Count
15
SkillsBench
51.1Pass Rate
12
SkillsBench (held-out)
53.3AUPRC
11
SkillsBench Real-task holdout n=65
74.2Set F1
10
SkillsBench Synthetic n=494 (test)
73.9Set F1
10
SkillsBench
22.6Pass Rate
9
SkillsBench 1000-skill scale
36.8Reward (%)
8
SkillsBench 1.0 (test)
71.1Pass Rate (With Skills)
8
SkillsBench n=81
25.9Verification Rate
7
SkillsBench
68.4Accuracy
6
SkillsBench
29.26Reward Mean (%)
6
SkillsBench 20-task balanced panel (held-out evaluation)
19Valid Task Count
5
SkillsBench 77 tasks (test)
0.46Mean Reward
4
SkillsBench 77 tasks
72.7Coverage
4
SkillsBench
105,000Solve Tokens
4
SkillsBench
39,000Tokens Used (Total)
4
SkillsBench
41,000Tokens Used
4
SkillsBench
52Solve Tokens (K)
4
SkillsBench
30,000Solve Tokens
4
SkillsBench
66Solve Tokens Count
4
SkillsBench
15Solve Tokens (K)
4
SkillsBench
22Baseline Success Rate
4