Frontier AI Risk Monitor - SOSBench-Chem: leaderboard
SOSBench's chemistry subset (500 regulation-grounded hazardous requests); an LLM judge decides refusal and the platform reverses the harmful-answer rate, so higher is safer.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 79 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.6 | 100 | #178 |
| 2 | GPT-5.5 | 99 | #49 |
| 3 | Hy4 preview | 99 | #68 |
| 4 | Claude Opus 4.8 | 98.8 | #56 |
| 5 | Qwen 3.5 Plus (Thinking) | 98.6 | #232 (Qwen 3.5 Plus) |
| 6 | Qwen 3.7 Max | 98.4 | #145 |
| 7 | GPT-6 Astra | 98.2 | #6 |
| 8 | Step 5 Preview | 97.4 | #108 |
| 9 | Grok 4.7 | 96.8 | #44 |
| 10 | GPT-5.6 Sol | 96.6 | #30 |
| 11 | Claude 3.5 Sonnet (20241022) | 95.8 | #637 |
| 12 | MiMo-V2.5-Pro | 95.6 | #279 |
| 13 | GLM-5.2 | 95.4 | #138 |
| 14 | GPT-5.4 (High) | 95.4 | #123 (GPT-5.4) |
| 15 | Grok 4.3 | 94.6 | #333 |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-sosbench-chem · How It Works · Data refreshed daily, snapshot 2026-11-13.