Benchmarks / Cyber
CWE-Bench
CWE-Bench: security-weakness detection; score.
CWE-Bench
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Grok 4.7 | xhigh | 68.3% |
| 2 | DeepSeek V4.1 Flash | max | 68.3% |
| 3 | GPT-6 Sol | max | 64.7% |
| 4 | GPT-6 Astra | max | 63.3% |
| 5 | MiMo V2.6 Pro | – | 63.3% |
| 6 | Muse Spark 1.3 | xhigh | 60.8% |
| 7 | Opus 5.5 | max | 58.3% |
| 8 | GPT-6 Luna | max | 56.7% |
| 9 | GLM 5.3 | max | 55.8% |
| 10 | GLM 5.3 Flash | max | 55.8% |
| 11 | Kimi K3 | max | 53.3% |
| 12 | Step 5 Preview | – | 52.5% |
| 13 | Gemini 3.8 Flash | high | 52.5% |
| 14 | Fable 5.1 | max | 50.8% |
| 15 | Mistral Large 4 | – | 50.8% |
| 16 | Beam 1.0 | – | 49.2% |
| 17 | MiniMax M3 | – | 41.7% |
| 18 | Qwen 3.8 2.4T-A95B | – | 28.3% |
| 19 | Nemotron 3 Ultra | max | 27.5% |
| 20 | Qwen 3.8 27B | xhigh | 26.7% |