Z.ai releases GLM-5.3 with coding and cybersecurity gains

Z.ai released GLM-5.3, built entirely through post-training on the GLM-5.2 base model. The model delivers a 50% improvement on the company’s internal Code Bench, open-source state-of-the-art results on Terminal Bench 3.0 and Agents’ Last Exam, and strong emergent cybersecurity performance including state-of-the-art scores on CyberGym for vulnerability discovery. Open weights are scheduled for release after safety evaluation.