According to CoinMeta, Zhipu's flagship model GLM 5.3 will be open-sourced at 10 a.m. on August 28th, Beijing time. The model was launched on August 14th with coding and plan, followed by the opening of API. Zhipu stated that during the post-training process of GLM 5.3, its network security capabilities improved faster than expected, enabling it to more effectively identify vulnerabilities and learn to plan a complete vulnerability exploitation chain across multiple steps. As a result of this, an official security assessment and reinforcement were conducted before deciding to make the weights of the model public. In the evaluation, GLM 5.3 scored 84.5% in the vulnerability discovery test Cybergym, slightly higher than Mythos 5's 83.8% and GPT-5.6 SOL's 83.6%. In the more attack-oriented Exploitbench test, GLM 5.3's score increased from 24.4% in GLM 5.2 to 54.4%, but it still lagged behind Mythos 5's 78.0%. Zhipu also mentioned that the model has identified 2,436 vulnerabilities in 269 open-source projects, of which 1,097 are considered severe or high-risk.