Cogent VR-1.1 beats Claude Code on vulnerability repair
Cogent Security has released VR-1.1, a reasoning model that writes fixes for newly disclosed software vulnerabilities. On the LiveCVEBench public benchmark covering 190 tasks across 14 programming languages, VR-1.1 produced working repairs for 57.9% of tasks, compared to 45.3% for both Claude Code and Codex. The model writes its own proof-of-concept exploits, traces related vulnerable code paths, reviews its own patches from a fresh context, and generates regression tests. Founded in 2025 and backed by $53 million in funding, Cogent says Fortune 500 customers have cut critical vulnerability exposure windows by 97%.
