Z.ai releases GLM-5.3
Z.ai releases GLM-5.3 with stronger long-horizon agent, coding and automated vulnerability-discovery capabilities.
What happened
Event details
Z.ai says GLM-5.3 extends post-training on the large GLM-5.2 base, focusing on long-horizon agent execution, verifiable coding workflows and automated vulnerability discovery. It is currently available through ZCode and GLM Coding Plan, with weights planned roughly two weeks later after security evaluation and hardening.
Assessment
Why it matters
GLM-5.3 applies large-scale post-training directly to long-horizon agent engineering and cybersecurity vulnerability discovery, while making the model's security evaluation part of the release process by delaying the weights.
Availability
Access notes
Available through ZCode and GLM Coding Plan; public weights and licensing remain pending completion of the security review.