CybersecurityMachine Learning4 min reading time

Cogent Reports VR-1.1 Passing Repairs on 57.9% of LiveCVEBench Tasks

Unite.AI
Read the full article
Cogent Security released VR-1.1, a vulnerability-repair model that successfully fixed 57.9% of tasks on the LiveCVEBench benchmark, outperforming Claude Code and Codex. VR-1.1 uses a security-focused training and iterative testing process to ensure fixes address vulnerabilities without breaking functionality.

More in Cybersecurity