[Zhipu Flagship Model GLM 5.3 to Be Open-Sourced at 10:00 AM on August 28] According to a report by Jinse Finance, the Zhipu flagship model GLM-5.3 will be open-sourced at 10:00 AM Beijing time on August 28. GLM-5.3 was actually launched on the Coding Plan platform as early as August 14, followed by the release of its API. Tomorrow, the model weights will finally be available for download, delayed by a full two weeks due to a rather unique reason. Zhipu stated that during the post-training process, the model's network security capabilities grew "beyond expectations," not only becoming better at identifying vulnerabilities but also learning to plan complete exploit chains across multiple steps. As a result, the team conducted security assessments and reinforcements before releasing the weights. In official evaluations, GLM-5.3 scored 84.5% in the Cyber Gym vulnerability detection test, slightly higher than Mythos 5's 83.8% and GPT-5.6Sol's 83.6%. In the more exploit-focused Exploit Bench test, its score jumped from GLM-5.2's 24.4% to 54.4%, though it still lags significantly behind Mythos 5's 78.0%. Zhipu also revealed that the model has identified 2,436 vulnerabilities across 269 open-source projects, 1,097 of which are classified as severe or high-risk. GLM-5.3 continues to use the same base model as GLM-5.2, with all these improvements stemming from post-training.