Collinear AI releases CWE-bench v2 benchmark

Collinear AI has released CWE-bench v2, a public benchmark featuring 100 tasks designed to test the ability of frontier AI models to find and fix security vulnerabilities in real open-source repositories. The update covers 91 different weakness classes and introduces a held-out set alongside updated leaderboards tracking open-weight and API-only models. Results show that leading models like Gemini 4 Argon and Claude Opus 5.5 solve a majority of the tasks, though a significant long tail remains unsolved. Open article →