GLM 4.5 Air

GLM 4.5 Air is a lighter member of Z.ai's GLM-4.5 family, engineered specifically for intelligent-agent applications where reasoning, coding, and tool use need to coexist. It uses a Mixture-of-Experts architecture with 106 billion total parameters and 12 billion active parameters, a more compact configuration than the flagship 355B/32B GLM-4.5. The design reflects a deliberate tradeoff: keeping expert-driven quality on complex tasks while lowering the inference cost and latency that agent loops demand.

A defining feature is its hybrid reasoning design, which lets a single deployment switch between a thinking mode for multi-step reasoning and tool orchestration and a non-thinking mode for immediate conversational responses, with reasoning toggled by an enable flag. Across twelve industry-standard benchmarks the GLM-4.5 series averaged 63.2, while GLM 4.5 Air delivered a competitive 59.8 with the efficiency advantages of its smaller active footprint, making it well suited for production agent pipelines, developer assistants, and retrieval or function-calling workflows that benefit from scalable reasoning on demand.