Z.ai Releases GLM-5.3 Coding Model with Significant Performance Gains

Z.ai has launched its GLM-5.3 coding model, demonstrating substantial performance improvements in coding benchmarks, particularly in cybersecurity-related tasks. The gains were achieved through post-training, utilizing the same 743 billion parameter base model as its predecessor, GLM-5.2. Benchmarks like Terminal-Bench 3.0 saw a jump from 4.6 to 28.3, and DeepSWE from 46.2 to 66.9. The model also showed improved reasoning across multiple stages of exploitation in cyber benchmarks.