HumanAgencyBench - Defer Important Decisions: leaderboard
HumanAgencyBench dimension Defer Important Decisions: 500 prompts testing whether the assistant appropriately leaves critical life decisions to the user rather than making them. 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. Saturation forecast: Not forecast. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 (20251101) | 88.2 | #81 |
| 2 | Claude Haiku 4.5 (20251001) | 87.08 | #264 |
| 3 | Claude Opus 4.7 | 86.28 | #51 |
| 4 | Claude Sonnet 4.6 | 83.24 | #94 |
| 5 | Gemini 2.5 Flash (Preview 04-17) | 80.48 | #246 |
| 6 | Claude Opus 4.6 | 75.96 | #67 |
| 7 | Claude 3.5 Sonnet (20240620) | 74.16 | #308 |
| 8 | Claude 3 Opus (20240229) | 69.68 | #428 |
| 9 | Claude 3.7 Sonnet (20250219) | 68.36 | #209 |
| 10 | GPT-5.4 | 64.76 | #83 |
| 11 | Claude 3.5 Haiku (20241022) | 60.6 | #590 |
| 12 | Claude 3.5 Sonnet (20241022) | 58.08 | #302 |
| 13 | Claude Opus 4.1 (20250805) | 57.84 | #139 |
| 14 | Claude Sonnet 4 (20250514) | 54.32 | #223 |
| 15 | GPT-5.5 | 49.92 | #33 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=humanagencybench-defer-important-decisions · How It Works · Data refreshed daily, snapshot 2026-11-05.