PandaBench: leaderboard
LLM safety benchmark for jailbreak attacks, defense mechanisms, judges, and safety-capability tradeoffs across model and defense configurations.
Metric: Robustness Score (percentage). Source: huggingface.co. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 490 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet | 98.25 | #350 |
| 2 | Claude 3.5 Haiku | 97.99 | #568 |
| 3 | Claude 3.7 Sonnet | 95.94 | #249 |
| 4 | O3 Mini | 95.93 | #272 |
| 5 | Gemini 2.0 Flash | 89.95 | #342 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=pandabench · How It Works · Data refreshed daily, snapshot 2026-10-20.