We have Mythos at Home: GLM 5.2 beats Claude in our Cyber Benchmarks
Summary
Semgrep compares open-weight GLM 5.2 against frontier models like Claude in IDOR vulnerability detection. The findings show GLM 5.2, run with no harness, scoring 39% F1 and outperforming Claude Code, while highlighting the cost advantage of open-weight models. The article emphasizes that the security harness and benchmarking context significantly influence results and discusses implications for security teams evaluating AI-powered code analysis.