Frontier AI Risk Monitor - SciKnowEval-BiologicalHarmfulQA - Red-Team Attacks: leaderboard
The same harmful biology questions wrapped in three FRT red-team attacks; the number is 100 minus the mean StrongREJECT harm score across the attacks.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 40 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 97 | #5 |
| 2 | Claude Opus 4.6 (Thinking) | 96.67 | #102 (Claude Opus 4.6) |
| 3 | Grok 4.7 | 95.33 | #44 |
| 4 | GPT-5.5 | 92.3 | #49 |
| 5 | Claude Opus 4.8 | 90.1 | #56 |
| 6 | GPT-6 Astra | 86.57 | #6 |
| 7 | GPT-5.6 Sol | 84.9 | #30 |
| 8 | Grok 4.3 | 83.03 | #333 |
| 9 | GPT-5.4 (High) | 73.47 | #123 (GPT-5.4) |
| 10 | Qwen 3.7 Max | 70.3 | #145 |
| 11 | Gemini 3.8 Flash | 65.75 | #36 |
| 12 | Qwen 3.5 Plus (Thinking) | 57.17 | #232 (Qwen 3.5 Plus) |
| 13 | GPT-5.1 (High) | 56.47 | #229 (GPT-5.1) |
| 14 | Kimi K2.6 | 55 | #178 |
| 15 | Hy4 preview | 52.4 | #68 |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-sciknoweval-biologicalharmfulqa-red-team-attacks · How It Works · Data refreshed daily, snapshot 2026-11-13.