Z.ai Releases GLM-5.3 With Major Coding and Cybersecurity Gains Through Post-Training

Z.ai has released GLM-5.3, a post-training refinement of the 743B-parameter GLM-5.2 base model that achieves significant performance gains in coding and cybersecurity. The model doubles long-horizon coding performance on Terminal-Bench 3.0, moving from 4.6 to 28.3, and reaches 84.5% on cybersecurity vulnerability discovery (CyberGym), ahead of competing models. Post-training involved expanded task environments and longer training sequences rather than a new base model. The model is accessible through the Z.ai API, GLM Coding Plan, and ZCode, with weights expected to be released approximately two weeks after launch following safety evaluations.