COMPASS Policy - Allowed Base: leaderboard
Organization-policy compliance across eight service domains under the base system prompt. Appendix Table 9, 15 models. Allowed/Denied and Base/Edge query types remain separate; higher policy alignment is better for all four. This is distinct from OpenCompass and TempCompass. This series reports Policy Alignment Score (%). Source: https://arxiv.org/html/2601.01836v1.
Metric: Policy Alignment Score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 99.78 | #105 |
| 2 | Gemma 3 12B (IT) | 99.65 | #672 |
| 3 | Gemini 2.5 Flash | 99.56 | #255 |
| 4 | Gemini 2.5 Pro | 99.18 | #159 |
| 5 | Gemma 3 4B (IT) | 98.84 | #997 |
| 6 | Llama 3.3 70B Instruct | 98.59 | #541 |
| 7 | Gemma 3 27B (IT) | 98.32 | #528 |
| 8 | Qwen 2.5 32B Instruct | 98.02 | #509 |
| 9 | Claude Sonnet 4 | 97.97 | #207 |
| 10 | Qwen 2.5 14B Instruct | 97.97 | #654 |
| 11 | Qwen 3 235B A22B Instruct | 97.84 | #336 |
| 12 | Qwen 2.5 7B Instruct | 97.69 | #863 |
| 13 | Qwen 2.5 72B Instruct | 97.57 | #469 |
| 14 | Kimi K2 | 97.5 | #252 |
| 15 | Gemma 3 1B (IT) | 92.04 | #1481 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=compass-policy-allowed-base · How It Works · Data refreshed daily, snapshot 2026-11-05.