Z.AI: GLM 4.5 Air

Z.AI's GLM-4.5-Air, a lightweight MoE model designed for agent-centric applications, demonstrates exceptional speed, consistently ranking among the fastest models across nine benchmarks. Its pricing is moderate, placing it in the 24th percentile. Reliability is a significant strength, with a 98% success rate indicating minimal technical failures. The model exhibits strong performance in several key areas. It achieves high accuracy in Reasoning (95.6%, 86th percentile), General Knowledge (99.5%, 75th percentile), and Mathematics (92.9%, 74th percentile), showcasing robust analytical and factual recall capabilities. Instruction Following also performs well at 68.7% accuracy (79th percentile). While its Hallucinations accuracy is 90.0%, placing it in the 35th percentile, it generally acknowledges uncertainty appropriately. A notable weakness is its Email Classification accuracy (80.0%, 9th percentile), suggesting room for improvement in nuanced categorization tasks. The model's "thinking mode" for advanced reasoning and tool use, alongside a "non-thinking mode" for real-time interaction, offers flexible deployment options.