Jailbreaking on HarmBench (Modern LLM Suite)
88Llama 3 8B ASRMetis
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MetisCategory=Multi-turn2026.05 | 88 | 90 | 97 | 86 | 94 | 93 | 76 | 78 | 90 | 100 | 89.2 | |
| X-TeamingCategory=Multi-turn2026.05 | 85 | 83 | 95 | 81 | 92 | 91 | 71 | 49 | 84 | 89 | 82 | |
| ActorBreakerCategory=Multi-turn2026.05 | 79 | 85.5 | 47 | 22 | 78.5 | 84.5 | 14 | 22 | 44 | 42 | 51.9 | |
| CrescendoCategory=Multi-turn2026.05 | 60 | 62 | — | — | 60 | 62 | 14 | — | 23 | 6 | 41 | |
| CodeAttackCategory=Single-turn2026.05 | 46 | 66 | 34 | 27 | 67 | 70.5 | 8 | 20 | 30 | 55 | 42.4 | |
| GCGCategory=Single-turn2026.05 | 34.5 | 17 | 6.5 | — | 55.8 | 12.5 | 0 | — | — | — | 21.1 | |
| CoACategory=Multi-turn2026.05 | 25.5 | 22.5 | 6 | 22 | 25.5 | 18.8 | 8 | 32 | 34 | 40 | 23.4 | |
| AutoDAN-TurboCategory=Single-turn2026.05 | 23 | 32 | 7 | 17 | 47 | 23 | 24 | 55 | 52 | 84 | 36.4 | |
| PAIRCategory=Single-turn2026.05 | 18.7 | 36 | 29.5 | — | 41 | 39 | 0 | — | — | — | 27.4 | |
| PAPCategory=Single-turn2026.05 | 16 | 16 | 31.5 | — | 40 | 42 | 0 | — | — | — | 24.3 | |
| CipherChatCategory=Single-turn2026.05 | 0 | 1.5 | 68 | 20 | 44.5 | 10 | 35 | 24 | 38 | 88 | 32.9 |