GLM‑4.5Open-source hybrid-reasoning MoE foundation model built for agentic, coding, and tool-use tasks

GLM-4.5 is an open-source large language model developed by Zhipu AI (Z.ai) as part of the GLM model family. It uses a Mixture-of-Experts (MoE) architecture and a hybrid-reasoning design that lets the model either "think" before responding or answer directly, targeting agentic workflows, coding, and tool use. The model supports a 128K-token context window and native tool calling. The model is positioned for developers building AI agents and coding assistants. It introduced "Interleaved Thinking," where the model reasons before each response and tool call, which later GLM releases (GLM-4.6 and GLM-4.7) extended with features like Preserved Thinking and Turn-level Thinking. GLM-4.5 emphasizes agentic coding, integrating with mainstream agent frameworks and coding tools such as Claude Code, Cline, Roo Code, and Kilo Code. The GitHub repository hosts model resources, inference code, and examples, while the weights are released openly for self-hosting and the API is offered through the Z.ai API Platform. The repository now also documents successor models GLM-4.6 (expanding context to 200K tokens) and GLM-4.7, alongside a lightweight 30B-A3B variant (GLM-4.7-Flash) for more efficient deployment. As an open-weight release, GLM-4.5 competes with other open models aimed at agentic and coding use cases. Its strengths lie in tool use, reasoning control, and openness, though running a large MoE model locally requires substantial hardware, and newer GLM versions have since superseded it on benchmarks.