Semgrep: GLM 5.2 beats Claude in our Cyber Benchmarks

Zhipu AI's GLM 5.2, an open-weight model, achieved a 39% F1 score on Semgrep's IDOR vulnerability detection benchmark, outperforming Claude Code (32%) at approximately $0.17 per vulnerability found. While Semgrep's proprietary multimodal pipeline still leads at 53-61% F1, GLM 5.2's strong performance using only a basic prompt harness—without specialized scaffolding—demonstrates that significant vulnerability-detection capability can come from the model itself rather than the surrounding infrastructure. The open-weight nature of GLM 5.2 makes it particularly valuable for security teams needing to run models in isolated environments.