Zhipu AI Releases GLM-5.3: Coding and Long-Horizon Task Capabilities Significantly Improved, Model Weights to Be Released in Two Weeks
TechFlow reports, on August 14, Z.ai released GLM-5.3, based on the same foundation model as GLM-5.2, achieving capability improvements through scaled post-training. Officials stated that GLM-5.3 improved by 50% over GLM-5.2 on the internal Z.ai Code Bench coding benchmark, and reached a leading position among open models in public benchmarks such as Terminal Bench 3.0 and Agents' Last Exam. In terms of cybersecurity, GLM-5.3 achieved a score of 84.5% in the CyberGym vulnerability discovery test, and significantly improved over the previous generation in exploitation chain-related tests such as ExploitBench and ExploitGym.
Z.ai stated that GLM-5.3 weights will be open two weeks after release, and security evaluation and hardening are currently still underway. In terms of API, the model no longer supports disabling thinking mode, and supports three levels of inference intensity: low, high, and max.





