Calibration
CWE-bench v1 → CWE-bench
CWE-bench is estimated from CWE-bench v1 with a Michaelis–Menten curve fitted on 6 models measured on both: y = 2.0000·x / (1.49229 + x), R² = 0.52, cross-validated error 6.6 pp. It is used for 2 estimates.
| Estimated model | CWE-bench v1 | CWE-bench | Source |
|---|---|---|---|
| Grok 4.6 | 57.0% | 55.3% | estimated ± 6.6 pp, low confidence |
| Muse Spark 1.3 | 55.0% | 53.9% | estimated ± 6.6 pp, low confidence |
Fitted on
| Model | CWE-bench v1 | CWE-bench |
|---|---|---|
| Claude Opus 5.5 | 67.0% | 67.1% |
| DeepSeek V4.1 Flash 1017 | 55.0% | 57.6% |
| Gemini 4 Argon | 68.0% | 65.0% |
| GPT-6 Astra | 68.0% | 64.2% |
| Grok 4.7 | 68.0% | 54.7% |
| Hy4 preview | 53.0% | 47.3% |