Frontier AI Risk Monitor - SafeSciBench-Chem - Red-Team Attacks: leaderboard
The same chemistry requests 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. 38 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 79.7 | #5 |
| 2 | Claude Opus 4.8 | 65.13 | #56 |
| 3 | Grok 4.3 | 56.17 | #333 |
| 4 | Claude Opus 4.6 (Thinking) | 54.91 | #102 (Claude Opus 4.6) |
| 5 | Grok 4.7 | 45.37 | #44 |
| 6 | GPT-5.5 | 41.6 | #49 |
| 7 | GPT-6 Astra | 41.3 | #6 |
| 8 | GPT-5.4 (High) | 40.53 | #123 (GPT-5.4) |
| 9 | Qwen 3.5 Plus (Thinking) | 34.67 | #232 (Qwen 3.5 Plus) |
| 10 | Gemini 3.8 Flash | 34.47 | #36 |
| 11 | Hy4 preview | 32.47 | #68 |
| 12 | MiniMax-M2 | 29.8 | #674 |
| 13 | GPT-5.1 (High) | 29.5 | #229 (GPT-5.1) |
| 14 | MiniMax-M2.7 | 27.47 | #518 |
| 15 | Gemini 3.1 Pro (Preview) | 26.5 | #94 |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-safescibench-chem-red-team-attacks · How It Works · Data refreshed daily, snapshot 2026-11-13.