HumanAgencyBench - Maintain Social Boundaries: leaderboard
HumanAgencyBench dimension Maintain Social Boundaries: 500 prompts checking whether the assistant keeps appropriate professional boundaries and avoids encouraging inappropriate personal relationships. o3-judged 0-10 rubric score averaged per model and shown on a 0-100 scale; higher is better.
Metric: HAB dimension score (0-100, o3-judged rubric). Source: github.com. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Haiku (20241022) | 93.52 | #590 |
| 2 | Claude 3.5 Sonnet (20240620) | 91.6 | #308 |
| 3 | Claude 3.5 Sonnet (20241022) | 89.22 | #302 |
| 4 | Claude Haiku 4.5 (20251001) | 88.3 | #264 |
| 5 | Claude Sonnet 4.6 | 85.02 | #94 |
| 6 | Claude Opus 4.6 | 67.42 | #67 |
| 7 | Claude Opus 4.5 (20251101) | 60.46 | #81 |
| 8 | Gemini 2.0 Flash | 58.34 | #351 |
| 9 | Claude Opus 4.7 | 56.9 | #51 |
| 10 | Claude 3.7 Sonnet (20250219) | 51.94 | #209 |
| 11 | Claude 3 Opus (20240229) | 48.3 | #428 |
| 12 | Gemini 1.5 Flash | 43.3 | #620 |
| 13 | Claude 3 Haiku (20240307) | 42.34 | #780 |
| 14 | Llama 4 Maverick Instruct | 40.46 | #467 |
| 15 | Kimi K2.6 | 38.88 | #113 |
Interactive version: theaggregate.ai/benchmark?slug=humanagencybench-maintain-social-boundaries · How It Works · Data refreshed daily, snapshot 2026-11-05.