ChatGLM
By Zhipu AI
ChatGLM is a series of bilingual (Chinese/English) conversational large language models developed by Zhipu AI, spun out of Tsinghua University's Knowledge Engineering Group (KEG) laboratory, built on the General Language Model (GLM)…
Definition
ChatGLM is a series of bilingual (Chinese/English) conversational large language models developed by Zhipu AI, spun out of Tsinghua University's Knowledge Engineering Group (KEG) laboratory, built on the General Language Model (GLM) architecture.
Overview
ChatGLM originated from academic research at Tsinghua University's KEG lab, which developed the GLM (General Language Model) pretraining architecture as an alternative to standard decoder-only transformers, using an autoregressive blank-filling objective that unifies language understanding and generation tasks. ChatGLM-6B, released in March 2023, was one of the first open, dialogue-tuned Chinese LLMs that individual developers could run on consumer GPUs, and it became widely adopted in the Chinese open-source AI community. Zhipu AI, the company commercializing this research, followed with ChatGLM2-6B and ChatGLM3-6B, each improving context length, reasoning, and tool-use capability while keeping the models runnable on modest hardware. In 2024, the company rebranded its flagship line as GLM-4, moving toward larger, more capable models offered through Zhipu's proprietary API (branded as Zhipu Qingyan/ChatGLM in consumer-facing products) while continuing to release smaller open-weight checkpoints such as GLM-4-9B for the community. Zhipu AI has grown into one of China's best-funded AI startups, sometimes grouped with Moonshot AI, MiniMax, 01.AI, Baichuan AI, and StepFun as the so-called "AI Tiger" companies competing in China's foundation model market. ChatGLM/GLM models are used both directly through Zhipu's consumer chat product and as a research and development base by academics and startups building Chinese-language applications, competing with Baichuan, InternLM, and Alibaba's Qwen in the open-weight space, and with proprietary offerings from Moonshot AI and MiniMax at the frontier end.
Key Concepts
- Built on the GLM (General Language Model) architecture using blank-filling pretraining
- ChatGLM-6B was among the first widely adopted open bilingual chat LLMs runnable on consumer GPUs
- Rebranded flagship line as GLM-4 with both proprietary and open-weight releases
- Strong optimization for low-resource, quantized local deployment
- Native bilingual Chinese/English dialogue tuning
- Tool-use and code-execution support added from ChatGLM3 onward
- Backed by Tsinghua University research lineage (KEG lab)