China’s GLM‑5.2 matches Anthropic’s Mythos in bug‑finding tests
Semgrep’s benchmark shows Zhipu AI’s open‑weight GLM‑5.2 model hit a 39% F1 on IDOR vulnerability detection, beating Anthropic’s Claude Code and nearing the performance of proprietary pipelines. At roughly one‑sixth the cost per bug, the result signals China’s home‑grown AI can now rival top U.S. models in cyber‑security tasks.