SkillsBench: leaderboard
SkillsBench (BenchFlow) scores agentic models on completing tasks across domains using modular Skills (folders of instructions, scripts, and resources), and also reports skill-lift, the with-minus-without-Skills gain.
Metric: With-skills score (%). Source: www.skillsbench.ai. Status: years away from saturation. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 67.3 |
| 2 | Claude Opus 4.7 | 61.2 |
| 3 | Gemini 3.1 Pro (Preview) | 60.8 |
| 4 | GLM-5.1 | 58.4 |
| 5 | Gemini 3 Flash | 54.6 |
| 6 | Claude Opus 4.8 | 54.1 |
| 7 | Kimi K2.6 | 54 |
| 8 | MiniMax-M3 | 53 |
| 9 | GPT-5.2 | 51.7 |
| 10 | Claude Opus 4.6 | 50.2 |
| 11 | DeepSeek V4 Pro | 50.1 |
| 12 | Claude Opus 4.5 | 49 |
| 13 | Gemini 3.5 Flash | 48.2 |
| 14 | Claude Sonnet 4.6 | 47.2 |
| 15 | DeepSeek V4 Flash | 44.7 |
Interactive version: theaggregate.ai/benchmark?slug=skillsbench · How It Works · Data refreshed daily, snapshot 2026-09-05.