GLM-4.5

GLM-4.5 is a new 355B parameter open-weight MoE model (32B active). It delivers state-of-the-art performance on reasoning, code, and agentic tasks. Both the 355B flagship and a 106B Air version are now available, featuring dual-mode inference.

GLM-4.5 is a cutting-edge open-weight MoE model designed to unify reasoning, coding, and agentic capabilities in a single framework. With 355 billion total parameters (32 billion active), it delivers state-of-the-art performance across diverse tasks, ranking third overall in benchmark comparisons. The model features dual-mode inference—thinking mode for complex problem-solving and non-thinking mode for quick responses—enhancing flexibility for different applications. Specializing in agentic tasks, GLM-4.5 supports 128K context length and native function calling, excelling in web browsing and coding scenarios. It outperforms rivals like Claude-4-Opus in BrowseComp benchmarks and achieves high tool-calling success rates. Additionally, it enables full-stack development, from frontend design to backend deployment, and integrates seamlessly with coding tools like Claude Code. Available via Z.ai’s platform, API, or locally through HuggingFace and ModelScope, GLM-4.5 combines advanced architecture—including MoE optimization and Muon training—with scalable reinforcement learning for robust performance. Its hybrid design and specialized post-training make it a versatile solution for complex AI-driven applications.