GLM 4.5 Air serves as the lightweight, agent-focused variant within the flagship model family. Built upon a Mixture-of-Experts architecture, it features a 106B-parameter design with 12B active parameters, allowing it to maintain high performance while significantly reducing resource requirements. This architecture is specifically engineered to support agent-centric workflows, providing a balance between computational efficiency and the capability to handle complex, multi-step tasks.

The model offers versatile operational modes, including a dedicated thinking mode for advanced reasoning and tool use, alongside a non-thinking mode optimized for real-time interaction. Users can toggle these behaviors to suit specific needs, such as coding, research, or creative projects. By combining this flexible inference approach with a large context window, the model is well-positioned for developers seeking a reliable, high-speed solution for building responsive AI agents.