GLM-5.3 Arrives: Same Base, Coding Up 50%, and Uncovers a 40-Year-Old World-Class Vulnerability
Zhipu AI has released GLM-5.3, an upgrade that maintains the original base architecture while delivering a 50% improvement in coding performance. Additionally, the model reportedly identified a critical security vulnerability that had been latent for 40 years.
Zhipu AI unveiled GLM-5.3, a new version of its large language model. Despite keeping the same underlying architecture as its predecessor, the model achieves a 50% boost in coding-related benchmarks. Beyond performance gains, the model demonstrated a remarkable capability by uncovering a major security vulnerability that had remained undetected for four decades. This discovery underscores the model's potential in cybersecurity analysis. The release has attracted attention from the tech community, highlighting Zhipu's progress in AI development.
Source: Zhipu GLM (GNews) —
original