The GLM-4.5-Air model is a specialized foundation model engineered specifically to support the complex requirements of intelligent agent applications. Built with a compact architecture, it features 106 billion total parameters and 12 billion active parameters, offering a streamlined alternative to its larger counterparts. The model is designed to unify reasoning, coding, and agentic workflows, providing users with a hybrid approach that includes both a dedicated thinking mode for complex problem-solving and tool usage, as well as a non-thinking mode for immediate, direct responses.

Developed as part of a broader series, this model benefits from a design lineage that emphasizes both performance and operational efficiency. It has demonstrated competitive results in industry-standard benchmarks, achieving a score of 59.8 while maintaining a smaller footprint than the flagship version. The model is available in various formats, including base and hybrid reasoning versions, with FP8 options provided to further optimize performance. Its design makes it a versatile choice for developers looking to integrate robust reasoning and agentic functionality into commercial projects or secondary development pipelines.