aiXamine - Safety Alignment: leaderboard
aiXamine Safety Alignment service score: recognizing and refusing harmful, unethical or policy-violating requests, scored with guard models and moderation APIs (aiXamine, arXiv 2608.20554v1, Appendix Table 7). Normalized 0-100 score aggregated from the service's tests; higher is better. 42 models evaluated under identical black-box conditions via OpenRouter or containerized vLLM.
Metric: Service score (0-100; normalized pass rate). Source: arxiv.org. 42 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Haiku | 99.3 | #571 |
| 2 | Phi-4 | 99 | #707 |
| 3 | cogito-v1-preview-qwen-14B | 98.4 | |
| 4 | GPT-OSS-120B | 98.2 | #346 |
| 5 | Gemma 4 E4B | 98.1 | #762 |
| 6 | Gemma 2 27B | 97.9 | #803 |
| 7 | Gemma 4 E2B | 97.7 | #1062 |
| 8 | GPT-4o | 97.5 | #357 |
| 9 | GPT-OSS-20B | 97.5 | #523 |
| 10 | Gemma 4 31B | 97.4 | #212 |
| 11 | DeepSeek R1 0528 | 97.4 | #232 |
| 12 | Gemma 4 26B A4B | 97.3 | #360 |
| 13 | Claude Haiku 4.5 | 97 | #287 |
| 14 | Llama 3.2 1B | 97 | #1540 |
| 15 | Grok 4 Fast | 97 | #266 |
Interactive version: theaggregate.ai/benchmark?slug=aixamine-safety-alignment · How It Works · Data refreshed daily, snapshot 2026-11-05.