ChatGLM2-6B is an open bilingual chat language model from Zhipu AI and Tsinghua’s KEG lab, and the second generation of the widely used ChatGLM-6B. Released as open weights on GitHub, where the repository has collected more than 15,000 stars, it was one of the first compact models to offer strong Chinese and English conversation quality that anyone could download and run locally.
The model has 6.2 billion parameters and is built on the GLM architecture with a hybrid training objective. It was pretrained on 1.4 trillion tokens of Chinese and English text, then aligned for dialogue with supervised fine-tuning and human feedback. Compared with the first generation, ChatGLM2-6B delivers noticeably stronger benchmark performance, a much longer 32K context window, and more efficient inference thanks to multi-query attention. Thanks to INT4 quantization, the chat model can run in as little as about 6 GB of GPU memory, which put local LLM deployment within reach of a single consumer graphics card. Code and scripts in the repository cover chat demos, CLI and web interfaces, fine-tuning, and deployment.
The model weights and code are free to download. The code is open source, and the model is released under a model license that permits commercial use subject to its terms, so there is no fee to Zhipu for using ChatGLM2-6B in your own projects. Your only costs are your own hardware or cloud compute.
ChatGLM2-6B is a good fit for developers and researchers who need a compact, bilingual open model for local experiments, Chinese-language chatbots, or as a fine-tuning base. It has since been succeeded by the GLM-4 family, but it remains a lightweight, well-documented starting point for learning how open chat models work. The repository’s last updates were in 2024.