GLM-5 is a new open-source large language model with 744B parameters (40B active) trained on 28.5T tokens, featuring DeepSeek Sparse Attention and a novel asynchronous RL infrastructure called slime. It achieves best-in-class performance among open-source models on reasoning, coding, and agentic tasks, ranking #1 on Vending Bench 2 for long-term operational capability. The model is released under MIT License and integrates with coding agents like Claude Code, OpenCode, and Cline, while also supporting document generation and cross-app automation through OpenClaw.
Sort: