CWE-bench
1 article in the last 30 days.
-
CWE-bench v2: More coverage, quality, and transparency
CWE-bench v2 publishes 100 tasks across 91 CWEs, where agents must find and fix security flaws in real repositories. Gemini 4 Argon leads the 11 evaluated models at 67.8% pass@1, and even the best model fails 19% of tasks.