> For the complete documentation index, see [llms.txt](https://brindha.gitbook.io/mylearning/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://brindha.gitbook.io/mylearning/tools/chatglm.md).

# ChatGLM

**What it is**

ChatGLM, developed by Zhipu AI and Tsinghua University, is a powerful family of large language models. The latest GLM-4 series significantly improves bilingual Chinese and English natural language processing, rivaling GPT-4 in performance and adaptability. 

**Architecture**

ChatGLM's architecture is based on the Generalized Language Model, originally designed for diverse NLP tasks. Its distinguishing feature is left-to-right autoregressive modeling combined with bidirectional encoding, allowing it to excel in both generative and discriminative tasks. The model uses FlashAttention, enabling it to handle up to 128K tokens for long-context tasks efficiently.

**GLM-4-9B capabilities**

GLM-4-9B-Chat has advanced features such as web browsing, code execution, custom tool calls, and long context reasoning supporting up to 128K context. This generation of models supports 26 languages including Japanese, Korean, and German. 

**Community scale**

Zhipu's open-source models like ChatGLM have been downloaded over 30 million times globally, earning more than 150,000 GitHub stars, making Zhipu the company with the most stars in China's large model field. 

**Commercial growth**

In 2024, Zhipu's commercial revenue more than doubled compared to 2023. Their platform is used by 700,000 enterprises and developers, with daily token consumption increasing 150 times and API annual revenue growing more than 30 times year-over-year. 

**Development timeline**

At the end of 2020, Zhipu developed the GLM pre-training architecture. By 2022, they developed and open-sourced the GLM-130B, a massive bilingual pre-training model. In 2023, they launched ChatGLM with hundreds of billions of parameters. In 2024, they released the next-generation GLM-4 with longer contexts, stronger multimodal capabilities, and faster inference.