Z.ai Releases GLM-5.3 Open-Weights Model with Advanced Coding and Cyber Capabilities

Z.ai has released GLM-5.3, an open-weights model that uses the same base model as GLM-5.2 but achieves significant performance gains purely through advanced post-training. The model achieves state-of-the-art results on coding and cybersecurity benchmarks, including Terminal Bench 3.0, Agents' Last Exam, and CyberGym. This release demonstrates that substantial capability improvements, particularly in complex coding, agentic tasks, and cybersecurity exploitation, can be unlocked solely through post-training optimization without retraining the base model. It provides the open-source community with a highly capable model for long-horizon digital labor and vulnerability discovery. GLM-5.3 shows a 50% improvement over GLM-5.2 on the Z.ai Code Bench and more than doubles its predecessor's performance on vulnerability exploitation benchmarks. Additionally, GGUF versions of the model have been made available via Unsloth for efficient local inference.