Z.ai Launches GLM-5.2 Model for Local Use with Advanced Features

GLM-5.2 is Z.ai's latest open model, featuring 744 billion parameters and a 1 million context window.

The model can be run locally using Unsloth Dynamic GGUFs, with quantization options that reduce memory usage while maintaining accuracy. For instance, the 2-bit dynamic quantization requires 239GB of disk space and can operate efficiently on systems with 256GB of RAM.

GLM-5.2 supports three thinking modes, allowing users to toggle between non-thinking and two levels of thinking for complex tasks. Recommended inference parameters include a temperature of 1.0 and top_p settings of 0.95 or 1.0.

The model's performance is comparable to other leading models, achieving approximately 76.2% accuracy with dynamic 1-bit quantization and 82% with dynamic 2-bit quantization.

Users can run GLM-5.2 in Unsloth Studio, which provides a user-friendly interface for managing models and settings. Installation instructions are provided for various operating systems, and the platform supports features like code execution and automatic parameter tuning.

Dynamic quantization techniques have been employed to optimize model performance while minimizing memory requirements, allowing for effective local deployment across different hardware configurations.