sf/zai-org/GLM-4.5-Air

GLM-4.5, released by Zhipu AI in July 2025, is its next-generation flagship large model. The model features a Mixture of Experts (MoE) architecture with a parameter scale of up to 355 billion, activating only a subset of experts during inference to achieve both efficiency and superior performance. GLM-4.5 supports an ultra-long context of 256K tokens and ranks first globally on the SuperCLUE leaderboard with a score of 89.3. It not only excels in Chinese language understanding and generation but also natively supports logical reasoning, code programming, and agent-based applications. With its openness and significantly lower API costs—just 30% of GPT-4's—it has become a leading example of domestically developed open-source large models.