Skill2-Bench - High Skill Entropy: leaderboard
Metric: Average per-step score (%; high skill-entropy tasks, the hardest skill switches; held-out test set of 300 cross-skill long-horizon tasks of 2 to 10 dependent steps over 558 skills in nine verifiable and open-ended domains, the whole task given in one prompt; verifiable steps graded for correctness, open-ended steps scored 0 to 1 by a Claude-opus-4.7 rubric judge; four samples, temperature 0.7 for open-weight models and provider defaults for API models). Source: arxiv.org. Saturation forecast: Around December 2026. 12 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 72.2 |
| 2 | GPT-5.5 | 71.9 |
| 3 | Claude Opus 4.7 | 68.6 |
| 4 | Gemini 3.1 Flash | 60.3 |
| 5 | Claude Sonnet 4.5 | 59.5 |
| 6 | O4 Mini | 58.6 |
| 7 | Claude Haiku 4.5 | 54.9 |
| 8 | GPT-5.4 Mini | 42.7 |
| 9 | Qwen 3 32B | 41 |
| 10 | Qwen 3 8B | 31.5 |
| 11 | Qwen 3 4B | 26 |
| 12 | OLMo 3 7B (Thinking) | 7.4 |
Interactive version: theaggregate.ai/benchmark?slug=skill2-bench-high-skill-entropy · How It Works · Data refreshed daily, snapshot 2026-09-26.