Frontier AI Risk Monitor - BioRiskBench: leaderboard
BioRiskBench (276 items) of biological-misuse requests judged by DeepSeek V4 Pro; the platform reports the reversed score as the model's safety level.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 41 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.8 Flash | 99.88 | #36 |
| 2 | Claude Opus 5.5 | 99.7 | #5 |
| 3 | Grok 4.7 | 99.52 | #44 |
| 4 | Grok 4.3 | 99.21 | #333 |
| 5 | Gemini 3.1 Pro (Preview) | 98.73 | #94 |
| 6 | GPT-6 Astra | 98.22 | #6 |
| 7 | Claude Opus 4.6 (Thinking) | 98.13 | #102 (Claude Opus 4.6) |
| 8 | Claude Opus 4.8 | 98.01 | #56 |
| 9 | GPT-5.5 | 97.25 | #49 |
| 10 | GPT-5.6 Sol | 96.17 | #30 |
| 11 | GPT-5.1 (High) | 92.33 | #229 (GPT-5.1) |
| 12 | Claude Opus 4.5 (Thinking) | 83.15 | #156 (Claude Opus 4.5) |
| 13 | Hy4 preview | 82.91 | #68 |
| 14 | Qwen 3.5 Plus (Thinking) | 81.49 | #232 (Qwen 3.5 Plus) |
| 15 | GPT-5.4 (High) | 79.98 | #123 (GPT-5.4) |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-bioriskbench · How It Works · Data refreshed daily, snapshot 2026-11-13.