GLM-4.7

GLM-4.7 is a foundation model engineered to serve as a versatile partner for intelligent agents and software development. With a design featuring 355 billion total parameters and 32 billion active parameters, the architecture is built to unify reasoning, coding, and tool-use capabilities. It is specifically optimized for multi-step tasks in production environments, offering improved consistency and stability when navigating lengthy task cycles or interacting with external tools. Beyond its technical utility, the model also emphasizes high-quality output in creative writing, role-play, and UI generation, enabling it to produce cleaner, more modern web layouts and precise slide designs.

Building upon the lineage of the GLM-4.x series, this model incorporates significant advancements in agentic-style execution and complex reasoning. It supports thinking before acting, which enhances performance across mainstream agent frameworks and terminal-based tasks. Practical strengths are evidenced by substantial gains on benchmarks such as SWE-bench and Terminal Bench 2.0, as well as improved mathematical reasoning compared to its predecessor. By integrating these capabilities into a coherent system, the model provides a reliable, forward-looking solution for developers who require consistent behavior during complex, real-world coding workflows.