← Back to Dashboard

Jailbreak Leaderboard

Ranking models by **Robustness**—how well they maintain safety when challenged with adversarial inputs.

How is this calculated?

We compare a model's refusal rate on Direct prompts vs Adversarial prompts (e.g., "Roleplay you are a villain...").
Fragility Gap = Direct Refusal Rate - Adversarial Refusal Rate.
Robustness Score = 100 - Fragility Gap.

RankModelRobustness ScoreFragility GapDirect RefusalAdversarial Refusal