Chinese GLM-5.2 Model Matches Leading U.S. AI Systems on Cybersecurity Benchmarks at Half the Cost

Z.ai released GLM-5.2 last week. The open-weight model matches the agentic capabilities of Claude Opus 4.8 and OpenAI's GPT-5.5 while costing roughly half as much to run, Axios reported. Two evaluations by Graphistry and Semgrep placed GLM-5.2 on par with leading U.S.

Models on cybersecurity investigation and vulnerability-discovery benchmarks. Graphistry researchers suggested the model may be an illegal distillation of both GPT-5.5 and Opus 4.8. Open-weight models allow direct download and modification.

Users can remove safety controls and fine-tune the system for specific tasks without a commercial provider. Graphistry stated GLM-5.2 is the first open-weight model it would recommend for a frontier-like cybersecurity experience. Hackers have discussed jailbreaking GLM-5.2 for hacking tasks in Russian-language forums, Jason Baker, managing security consultant at GuidePoint Security, told Axios.

Travis Lanham, CTO and founder of Armadin, said the model lets attackers personalize operations after initial access by finding ways to move laterally and chain exploits. Roye Bass, a ransomware threat intelligence analyst at Halcyon, said GLM-5.2 removes another barrier for hackers who buy purpose-built malicious LLMs and stolen API keys.

Attackers can now download the model, run it locally, and generate phishing emails or fraud scripts themselves.

Z.ai did not respond to a request for comment. Z.ai founder Jie Tang has said publicly that the company will likely release an open-source model rivaling Anthropic's Fable before the end of the year. 360 Technology said this week that it has developed its own version of Mythos.