Secure Code Generation on CyberSecEval SCG
79.06SafetySafeCoder
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SafeCoderBackbone=CodeLlama-7B2026.02 | 79.06 | 25.85 | 18.7 | |
| ProSecBackbone=CodeLlama-7B2026.02 | 77 | 13.91 | 9.64 | |
| PurpCode-14B2026.02 | 76.51 | 46.37 | 32.34 | |
| SecCoderXBackbone=CodeLlama-7B2026.02 | 73.56 | 38.64 | 25.81 | |
| ProSecBackbone=Qwen2.5-Coder-3B2026.02 | 71.65 | 22.65 | 14.35 | |
| SafeCoderBackbone=Qwen2.5-Coder-3B2026.02 | 70.24 | 35.59 | 22.53 | |
| SecCoderXBackbone=Qwen2.5-Coder-7B2026.02 | 69.4 | 56.53 | 37.32 | |
| SecCoderXBackbone=Qwen2.5-Coder-7B2026.02 | 69.4 | 56.53 | 37.32 | |
| ProSecBackbone=Qwen2.5-Coder-7B, Reproduced=true2026.02 | 69.25 | 25.66 | 16.06 | |
| SecCoderXBackbone=Qwen2.5-Coder-3B2026.02 | 68.7 | 45.45 | 29.5 | |
| SafeCoderBackbone=Qwen2.5-Coder-7B2026.02 | 67.93 | 39.38 | 24.34 | |
| BaseBackbone=CodeLlama-7B2026.02 | 64.45 | 36.29 | 20.96 | |
| Gemini-2.5-Flash2026.02 | 63.09 | 95.14 | 59.68 | |
| BaseBackbone=Qwen2.5-Coder-3B2026.02 | 62.38 | 45.59 | 26.49 | |
| BaseBackbone=Qwen2.5-Coder-7B2026.02 | 61.53 | 54.33 | 31.98 | |
| Qwen2.5-Coder-14B-Inst2026.02 | 61.37 | 66.73 | 40.16 | |
| GPT-4.12026.02 | 61.32 | 88.11 | 53.85 |