GLM-5.3 Emerges as a Cyber Powerhouse, Doubling Exploit Capabilities

GLM-5.3: Frontier Coding with Emergent Cyber Capabilities

Z.ai's GLM-5.3, built solely through post-training scaling, achieves state-of-the-art coding and cyber capabilities. It doubles exploitation benchmark scores over GLM-5.2, excelling in vulnerability discovery and complex long-horizon tasks. The model identified 2,436 real-world vulnerabilities across 269 projects, some dormant for decades. Weights will be open-sourced after safety checks.