On March 14, 2023, Tsinghua University’s Knowledge Engineering Group (KEG) and the spin-out company Zhipu AI open-sourced ChatGLM-6B, an open bilingual (Chinese-English) conversational model with 6.2 billion parameters built on their General Language Model (GLM) framework. It was trained on roughly a trillion tokens and tuned with supervised fine-tuning, feedback bootstrapping, and reinforcement learning from human feedback.
The standout feature was accessibility. With INT4 quantization, ChatGLM-6B needed only about 6 GB of GPU memory, letting it run on a consumer-grade graphics card rather than a data-center cluster. The weights were opened for academic use and, after registration, for free commercial use. The team’s larger GLM-130B model had been released in August 2022, and ChatGLM-6B brought that work to a far wider audience.
ChatGLM-6B was downloaded heavily - the GLM open models reported over 10 million Hugging Face downloads in 2023 - and the family advanced quickly through ChatGLM2, ChatGLM3, and the GLM-4 series. It helped make local, low-resource deployment of capable chat models practical and put Zhipu AI among China’s leading model labs.