Code Generation on MAB code
69.6TSw/o Attack
Evaluation Results
| Method | Links | |
|---|---|---|
| w/o AttackArchitecture=Hier.2026.05 | 69.6 | |
| Web FraudArchitecture=Hier.2026.05 | 67.8 | |
| w/o AttackArchitecture=Flat2026.05 | 66.2 | |
| Prompt InfectionArchitecture=Hier.2026.05 | 65.4 | |
| InjecAgentArchitecture=Hier.2026.05 | 63.9 | |
| w/o AttackArchitecture=Chain2026.05 | 63.4 | |
| Web FraudArchitecture=Flat2026.05 | 62.4 | |
| Web FraudArchitecture=Chain2026.05 | 61.8 | |
| InjecAgentArchitecture=Flat2026.05 | 59.2 | |
| Forced OutputArchitecture=Hier.2026.05 | 58.3 | |
| InjecAgentArchitecture=Chain2026.05 | 57.6 | |
| Prompt InfectionArchitecture=Chain2026.05 | 54.8 | |
| Prompt InfectionArchitecture=Flat2026.05 | 51.6 | |
| Evo-AttackerArchitecture=Hier.2026.05 | 49.7 | |
| Forced OutputArchitecture=Flat2026.05 | 48.8 | |
| Forced OutputArchitecture=Chain2026.05 | 47.9 | |
| Evo-AttackerArchitecture=Flat2026.05 | 39.2 | |
| Evo-AttackerArchitecture=Chain2026.05 | 37.8 |