Jailbreaking on AdvBench (Contemporary LLM Performance)
33.5GPT-3.5 ASRGCG
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| GCGCategory=Single-turn2026.05 | 33.5 | 21 | 12 | 13 | — | 12.5 | — | — | — | — | 18.4 | |
| PAPCategory=Single-turn2026.05 | 36 | — | — | 26.5 | — | 24.5 | — | — | — | — | 29 | |
| CipherChatCategory=Single-turn2026.05 | 41.5 | 61 | 31 | 55 | 0 | 32 | 1 | 72 | 37 | 63 | 39.4 | |
| AutoDAN-TurboCategory=Single-turn2026.05 | 44 | — | — | 4 | 6 | 62 | 40 | 48 | 36 | 57 | 37.1 | |
| CodeAttackCategory=Single-turn2026.05 | 44.5 | 64 | — | 8 | 18 | 56 | 9 | 23 | 10 | 69 | 33.5 | |
| ActorAttackCategory=Multi-turn2026.05 | 47.5 | — | — | — | 44 | 51.5 | 30 | 50 | 74 | 60 | 51 | |
| CoACategory=Multi-turn2026.05 | 52 | — | — | — | 30 | 63.5 | 30 | 0 | 72 | 62 | 44.2 | |
| PAIRCategory=Single-turn2026.05 | 57.5 | 20.7 | 52 | 25 | — | 61 | — | — | — | — | 43.2 | |
| MetisCategory=Multi-turn2026.05 | 98 | 88 | 94 | 100 | 84 | 96 | 60 | 90 | 48 | 100 | 85.8 |