Z.ai's GLM-5.2 is forcing security teams to take open-weight models seriously for vulnerability discovery, code review, and agentic security work. The practical question is no longer whether open models can compete, but how teams should evaluate them safely.