Zai releases GLM-5.3 with post-training gains in coding and cyber

Zai has released GLM-5.3, an open-weights model that retains the base architecture of GLM-5.2 but achieves significant performance improvements through post-training.

• It is described as the most capable open-weights coding model, showing a 50% improvement over GLM-5.2 on Z.ai Code Bench and achieving open-source SOTA on Terminal Bench 3.0 and Agents' Last Exam.
• The model exhibits emergent cyber capabilities, setting state of the art on CyberGym for vulnerability discovery and more than doubling GLM-5.2's performance on exploitation benchmarks.

The release provides updated checkpoints for users requiring advanced coding and security analysis capabilities.