Frontier AI Risk Monitor - AirBench-SecurityRisks - Red-Team Attacks: leaderboard
The same AIR-Bench security-risk requests wrapped in the platform's FRT red-team attacks; the number is the share of attacks that did not succeed.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 44 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.7 | 59 | #44 |
| 2 | Claude Opus 5.5 | 54 | #5 |
| 3 | GPT-5.5 | 47.67 | #49 |
| 4 | GPT-5.4 (High) | 43.17 | #123 (GPT-5.4) |
| 5 | Grok 4.3 | 38.17 | #333 |
| 6 | Hy4 preview | 38 | #68 |
| 7 | Claude Opus 4.5 (Thinking) | 37 | #156 (Claude Opus 4.5) |
| 8 | GPT-6 Astra | 36.5 | #6 |
| 9 | GPT-5.6 Sol | 34.33 | #30 |
| 10 | Claude Opus 4.8 | 33.83 | #56 |
| 11 | MiniMax-M3 | 33.83 | #269 |
| 12 | Qwen 3.5 Plus (Thinking) | 32.17 | #232 (Qwen 3.5 Plus) |
| 13 | Step 5 Preview | 30 | #108 |
| 14 | Claude Opus 4.6 (Thinking) | 28.5 | #102 (Claude Opus 4.6) |
| 15 | MiniMax-M2.7 | 26.67 | #518 |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-airbench-securityrisks-red-team-attacks · How It Works · Data refreshed daily, snapshot 2026-11-13.