“ZAI's GLM-5.3 achieved state-of-the-art results on the AutomationBench and GDPVal benchmarks for agentic use cases, slightly outperforming US frontier models.”