Z.ai releases GLM-5.3 with improved coding and cybersecurity via post-training

Z.ai has released GLM-5.3, an update to its 743B parameter model that achieves performance gains through scaled post-training rather than base model retraining. The release is currently available via the Z.ai API, GLM Coding Plan, and ZCode, with public weights scheduled for release approximately two weeks after launch following safety evaluations.

• Terminal-Bench 3.0 scores increased from 4.6 to 28.3, while DeepSWE v1.1 improved from 46.2 to 66.9.
• CyberGym results reached 84.5%, surpassing Mythos 5 (83.8%) and GPT-5.6 Sol (83.6%).
• ExploitBench scores more than doubled to 54.4%, though they remain below Mythos 5's 78.0%.
• On Z.ai Code Bench, the model reported a 50% improvement over GLM-5.2, scoring 31.4% at roughly 50,000 output tokens per task.

The update is positioned for developer tooling, application security, and fintech engineering, with specific applications including repository-scale refactors, long-horizon CLI agents, and white-box vulnerability discovery.