Leaderboard
Pass rate by condition.
Preliminary · 9 paired runs · October 27, 2026 · Education tasks (3)
Pass rate · With NOUS skills
77.2%
Pass rate · No skills
70.2%
Difference
+7.0 pts
Pass rate by model and condition
| Model | No skills | NOUS skills | Self-generated skills | Paired runs |
|---|---|---|---|---|
| Claude Haiku 4.5 | 100.0% | 90.0% | Not yet run | 1 |
| Gemma 4 | 67.1% | 72.5% | Not yet run | 4 |
| gpt-oss 20B | 66.7% | 78.8% | Not yet run | 4 |
Model labels reflect repository defaults rather than independently confirmed model IDs for every recorded run.
6
NOUS scored higher
2
Baseline scored higher
1
Tie
Models (3/3)
fleet 80.4%
Claude Haiku 4.5
90.0%
Gemma 4
72.5%
gpt-oss 20B
78.8%
0%20%40%60%80%100%
Preliminary · 9 paired runs · October 27, 2026 · Education workflows. Founder-reported. The dashed line marks the mean of the selected models. Scores are averages of automated quality checks, not success rates across all agent tasks; model labels use repository defaults.