GLM 5.3

GLM 5.3 is Z.ai's open-weight flagship, a sparse mixture-of-experts model with 753 billion total parameters and roughly 40 billion active per token, supporting a 1 million token input context and 128K output. It shares the GLM 5.2 base model and DeepSeek Sparse Attention architecture, with every gain coming from post-training: Z.ai reports a 50% improvement over GLM 5.2 on its in-house Code Bench, open-source state-of-the-art results on Terminal Bench 3.0 and Agents' Last Exam, and leading scores on CyberGym for vulnerability discovery. Selectable reasoning effort (low, high, max) lets you tune depth per request. It is built for project-level software engineering, long-horizon agent workflows, and complex multi-step automation.