Frontier AI Risk Monitor - AirBench-SecurityRisks: leaderboard
AIR-Bench prompts asking for cyber-attack help (100 items), sent without attack; the number is the platform's reversed harm score, so higher is safer. The Red-Team Attacks board sends them through the FRT jailbreak framework.
Metric: Safety score (0-100, higher is safer). Source: airiskmonitor.net. 44 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.7 | 100 | #44 |
| 2 | MiniMax-M3 | 99.5 | #269 |
| 3 | MiniMax-M2 | 99.5 | #674 |
| 4 | MiniMax-M2.7 | 99 | #518 |
| 5 | Claude Opus 5.5 | 99 | #5 |
| 6 | Qwen 3.5 Plus (Thinking) | 98.5 | #232 (Qwen 3.5 Plus) |
| 7 | GPT-5.4 (High) | 98 | #123 (GPT-5.4) |
| 8 | GPT-5.5 | 97 | #49 |
| 9 | Claude Opus 4.5 (Thinking) | 96.5 | #156 (Claude Opus 4.5) |
| 10 | GPT-5.6 Sol | 96 | #30 |
| 11 | Kimi K2.6 | 95.5 | #178 |
| 12 | Hy4 preview | 94.5 | #68 |
| 13 | Grok 4.3 | 93.5 | #333 |
| 14 | MiMo-V2.5-Pro | 90.5 | #279 |
| 15 | GPT-5.1 (High) | 90 | #229 (GPT-5.1) |
Interactive version: theaggregate.ai/benchmark?slug=frontier-ai-risk-monitor-airbench-securityrisks · How It Works · Data refreshed daily, snapshot 2026-11-13.