Frontier AI Risk Monitor - Fortress-Chemical: leaderboard
Scale's FORTRESS adversarial prompts in the chemical subdomain (37 prompts, 5 epochs) graded against per-prompt rubrics; the platform reverses the risk score, so higher is safer.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 46 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 | 98.61 | #5 |
| 2 | Claude Opus 4.5 (Thinking) | 97.45 | #156 (Claude Opus 4.5) |
| 3 | Gemini 3.8 Flash | 95.29 | #36 |
| 4 | Grok 4.7 | 94.75 | #44 |
| 5 | Claude Opus 4.8 | 91.89 | #56 |
| 6 | GPT-5.4 (High) | 91.74 | #123 (GPT-5.4) |
| 7 | GPT-5.5 | 90.08 | #49 |
| 8 | GPT-6 Astra | 89.36 | #6 |
| 9 | Claude Opus 4.6 (Thinking) | 87.41 | #102 (Claude Opus 4.6) |
| 10 | GPT-5.6 Sol | 83.44 | #30 |
| 11 | Grok 4.3 | 80.85 | #333 |
| 12 | MiniMax-M2.7 | 79.77 | #518 |
| 13 | MiniMax-M2 | 79.09 | #674 |
| 14 | Qwen 3.5 Plus (Thinking) | 78.92 | #232 (Qwen 3.5 Plus) |
| 15 | MiMo-V2.5-Pro | 74.05 | #279 |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-fortress-chemical · How It Works · Data refreshed daily, snapshot 2026-11-13.