GLM-4 9B Chat

The GLM family comes out of Tsinghua University's Knowledge Engineering Group and Zhipu AI. GLM-4-9B-Chat is the open-weight 9-billion-parameter member of the GLM-4 series — the same series that powers Zhipu's hosted ChatGLM service in China.

What stands out

For its size, GLM-4-9B-Chat is unusually competitive with much larger models on instruction-following and bilingual (Chinese/English) tasks. The 128K context variant is also notable — most 9B-class models still cap out at 32K.

What it's good at

Chinese-language tasks, JSON-mode structured output, function calling, and long-document reasoning at modest hardware cost. Popular for self-hosted bilingual assistants in academic and small-business settings.

Variants

The chat checkpoint comes in 8K, 128K, and 1M context lengths. The 1M variant uses a sliding-window attention approach and is mostly useful for needle-in-haystack retrieval rather than coherent generation over the full context.

License

GLM-4 Model License — permissive for non-commercial use and most commercial use, with a notification requirement for large-scale deployment. Read the full license before commercial deployment.