Security-aware code generation on LLMSecEval+
40Functional Success @1Reprompt
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RepromptModel=AR2026.05 | 40 | 64.7 | 26 | 40 | 48.7 | |
| VanillaModel=AR2026.05 | 37.3 | 55.3 | 21.3 | 30.7 | 38 | |
| Sec promptModel=Diffusion2026.05 | 34.7 | 58.7 | 16 | 22.7 | 26.7 | |
| VanillaModel=Diffusion2026.05 | 32 | 54.7 | 14.7 | 22 | 24 | |
| CDCModel=Diffusion2026.05 | 30.7 | 80.7 | 24.7 | 40.7 | 45.3 |