Z.ai launches GLM-5.3, boosting coding and cybersecurity prowess

Z.ai’s new GLM-5.3 model excels in coding and cybersecurity, achieving record benchmarks and uncovering thousands of vulnerabilities.

GLM-5.3 sets new records for open-source AI models, outperforming its predecessor GLM-5.2 by 50% on coding benchmarks.
The model has identified over 2,400 vulnerabilities across 269 software projects, with many flaws rated medium or higher in severity.
Z.ai utilized advanced training techniques, including specialized sandboxes, to enhance the model’s long-horizon coding capabilities.