Chinese startup Zhipu AI has released GLM-5.3, an upgraded coding model based on GLM-5.2, claiming it is now the strongest open-weights system for programming and complex, agent-style problem solving. Trained on cybersecurity-focused data and environments, GLM-5.3 can reason through multi-stage software exploitation, building coherent end-to-end attack chains that expose weaknesses in legacy and contemporary codebases spanning decades. Working with Chinese security teams, Zhipu AI reports discovering 2,436 vulnerabilities across 269 projects, documented in a public registry. GLM-5.3 is already available via subscription, with model weights slated for open release after security review.
This update represents a notable development in the Ai sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.