We have Mythos at Home: GLM 5.2 Beats Claude in Our Cyber Benchmarks
Summary
Semgrep’s blog reports on a benchmark comparing vulnerability-detection models for IDOR tasks. GLM 5.2, an open-weight model from Zhipu AI, beat Claude Code in a prompt-only setup, but Semgrep's multimodal harness still leads. The piece emphasizes harness quality over model alone and notes GLM 5.2’s relatively low cost (~$0.17 per vulnerability) and the potential for open-weight models to compete with frontier LLMs, while cautioning that results are dataset- and task-specific.