Z.ai launches GLM-5.3 AI model with enhanced coding capabilities

Z.ai, the international brand of Chinese AI company Zhipu, has released GLM-5.3, its latest AI model update. The new version is enhanced for performance in coding tasks, handling long-horizon projects, and improving cybersecurity capabilities.

The company states that the performance gains in GLM-5.3, which shares its base model with GLM-5.2, are attributed to post-training refinements. Z.ai reported that GLM-5.3 achieved a 50% increase in score on its internal Z.ai Code Bench benchmark compared to its predecessor. The model also reportedly reached leading open-source results on Terminal-Bench 3.0 and Agents’ Last Exam.

Furthermore, GLM-5.3 demonstrated improved performance in vulnerability discovery and exploitation within cybersecurity. Z.ai plans to release the model weights two weeks after the initial launch, following additional safety evaluations. Specific details on the extent of the cybersecurity improvements and the methodology for the benchmark testing have not been fully disclosed.