According to CoinMeta, Zhispu has officially released GLM-5.3, which builds upon the foundation of GLM-5.2. All improvements come from post-training, with a focus on enhancing programming skills, long-term tasks agent, and network security. Programming capabilities have seen a significant improvement, with terminal-bench rising from 4.6 to 28.3 and deepswe from 46.2 to 66.9. In terms of network security, cybergym achieved a score of 84.5, surpassing GPT-5.6's score of 83.6. Zhispu also stated that since GLM-5.2, they have identified 2,436 vulnerabilities in 269 open-source projects in collaboration with their security team, of which 53 have been publicly disclosed. The model weights will be made public within two weeks after completing the security assessment and reinforcement process.