Z.AI releases GLM-5.3, an open-weight coding model

China-based Z.AI announced the release of GLM-5.3, an open-weight large language model aimed at code generation. The company said the update delivers higher performance on standard coding benchmarks compared with publicly available models.

According to Z.AI, changes to the training and fine-tuning pipeline target improvements in code completion, code synthesis and debugging. The company noted stronger synthetic reasoning on programming problems and a reduction in common code-generation errors versus earlier versions.

The release makes GLM-5.3’s parameters available for researchers and developers to inspect, adapt and deploy. Z.AI described the openness as a way to accelerate experimentation, allow fine-tuning for specific use cases and support on-premise deployment in enterprise settings.

Z.AI also reported steps to limit unsafe or harmful outputs in coding contexts. The model joins other open-weight coding-focused projects from academic labs and companies that publish weights for community use.

Third parties can run independent benchmarks to test GLM-5.3’s code generation on real-world software tasks and inspect its training characteristics. Z.AI presented the release as part of its effort to expand publicly accessible models for developer workflows, including code completion and automated testing assistance.