Cogent Security releases VR-1.1 vulnerability repair model

Vulnerability management startup Cogent Security has released VR-1.1, a new frontier reasoning model that writes fixes for newly disclosed software vulnerabilities. On the LiveCVEBench benchmark, VR-1.1 successfully produced working repairs for 57.9 percent of tasks, outscoring Anthropic's Claude Code and OpenAI's Codex, which tied at 45.3 percent. The model is designed to check its own work for security gaps and functionality breaks before finishing, and it leaves behind regression tests. Open article →