H: Zhipu AI will release GLM-5.3 within 90 days, focused on improving safety/harm rate performance (tar

The sentiment trajectory shows decelerating positive sentiment and a recent negative shift. The most common criticism of Chinese open-source models is safety alignment. Claude Opus 4.8 achieved 2.5% harm rate on WorkBench — this is the new competitive frontier. Zhipu needs to address this to maintain enterprise credibility.