When Lower Privileges Suffice: leaderboard
As LLM agents increasingly select tools autonomously, their choices among tools with different privileges become safety-relevant. However, prior tool-selection studies focus on safety-agnostic metadata preferences, leaving privilege-sensitive choices underexplored. To address this gap, we study over-privileged tool selection, in which an agent selects or escalates to a higher-privilege tool despit
Metric: Business (self-reported). Source: benchmarklist.com. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 8B | 61.8 | #674 |
| 2 | Llama 3.1 8B | 47.4 | #1151 |
| 3 | MiniMax-M2.7 | 40.8 | #258 |
| 4 | Qwen 3.5 397B A17B | 36.8 | #146 |
| 5 | DeepSeek V3.2 | 32.9 | #205 |
| 6 | Grok 4.1 Fast | 30.3 | #217 |
| 7 | Kimi K2.5 | 17.8 | #144 |
| 8 | Gemini 3 Flash | 13.2 | #99 |
| 9 | GPT-5.2 | 9.2 | #110 |
| 10 | GLM-5 | 5.3 | #141 |
| 11 | Claude Sonnet 4.6 | 0 | #90 |
Interactive version: theaggregate.ai/benchmark?slug=when-lower-privileges-suffice · How It Works · Data refreshed daily, snapshot 2026-10-20.