China's Z.ai releases open-weight GLM-5.3, claims parity with Anthropic on cyber defense benchmarks
China's Z.ai released open-weight GLM-5.3 Friday, saying it edges Anthropic's closed Mythos 5 on vulnerability detection: 84.5% vs 83.8% (CyberGym benchmark). Offensive functions stay locked to verified users. Beijing's clearest bid yet for open AI as a cyber-defense default.
Chinese AI developer Z.ai released its new open-weight model GLM-5.3 on Friday, positioning it explicitly as a cyber-defense tool. The company says GLM-5.3 scored 84.5% on the CyberGym benchmark, which tests a model's ability to review code and find security flaws, edging out Anthropic's closed-source Mythos 5 at 83.8%. Offensive capabilities such as exploit development remain restricted to a verified 'Cybersecurity Trusted Access' program, while defensive functions like code review and vulnerability scanning are openly available. The release matters because it lowers the barrier for defenders anywhere, not just companies that can afford closed frontier models from Anthropic or OpenAI, to run capable security-analysis AI locally. It is also the clearest example yet of Chinese AI labs using open weights as a strategic tool to compete with better-funded US labs on a specific, high-value use case rather than general-purpose capability alone.