GLM-5.3 ties top open models via post-training; release paused for safety

DeepLearningAI · x · 2026-08-31

GLM-5.3 scored 60 on the Artificial Analysis Intelligence Index, tying for the top spot among open weights models. Its gains came entirely from fine-tuning GLM-5.2 rather than a new base architecture. Training inside software engineering environments led to emergent cybersecurity capabilities (84.5% on CyberGym). The model's release is temporarily paused for security evaluation due to unexpected exploit generation skills.

Original post →

More from Models

Models channel →