U.S. outlets The Verge and the Wall Street Journal report Zhipu (02513.HK) has released open-weight model GLM-5.2. The model still lags Anthropic and OpenAI on broad tasks, but some researchers say GLM-5.2 matches Anthropic’s Mythos in certain vulnerability-detection and cybersecurity scenarios. Cybersecurity firm Semgrep reports GLM-5.2 outperformed Anthropic’s Claude Opus 4.8 on some benchmarks; researchers add that with additional instruction both Opus 4.8 and GLM-5.2 can rival Mythos at bug

U.S. outlets The Verge and the Wall Street Journal report Zhipu (02513.HK) has released open-weight model GLM-5.2. The model still lags Anthropic and OpenAI on broad tasks, but some researchers say GLM-5.2 matches Anthropic’s Mythos in certain vulnerability-detection and cybersecurity scenarios. Cybersecurity firm Semgrep reports GLM-5.2 outperformed Anthropic’s Claude Opus 4.8 on some benchmarks; researchers add that with additional instruction both Opus 4.8 and GLM-5.2 can rival Mythos at bug finding. GLM’s open-weight distribution — downloadable and runnable on commodity hardware — boosts flexibility and deep access for advanced users but may elevate the risk of malicious misuse.