GLM-5.3 Official ReleaseProvider report
2026-08-14Based on the complete model-comparison table on Zhipu AI's official GLM-5.3 release page dated August 14, 2026. The source contains eight models and 16 benchmark rows. This site splits the paired 2-hour / 6-hour ExploitGym cells into 17 metrics and 120 available scores. Fable 5's fallback marker and every original model-column label are retained in raw; null cells do not create scores.