ChatGLM3-6B is an open-source, bilingual language model developed jointly by Zhipu AI and Tsinghua University's KEG Lab. As the third generation of the ChatGLM series, it is built upon the General Language Model (GLM) architecture and optimized for both Chinese and English processing. The model is designed to be efficient enough for deployment on consumer-grade graphics cards while maintaining high performance across various reasoning and semantic benchmarks.
One of the most significant upgrades in ChatGLM3-6B is its native support for complex functionalities, including function calling, code interpretation, and agent-based tasks. It utilizes a newly designed prompt format that enables more structured interactions, allowing the model to interact with external tools and execute Python code to solve mathematical or data-driven problems.
The model incorporates technical features such as Rotary Positional Embeddings (RoPE) and a more diverse training dataset compared to its predecessors. It was released alongside a suite of related models, including a base version for fine-tuning and a specialized long-context variant. ChatGLM3-6B supports a standard context window of 8,192 tokens, providing sufficient memory for multi-turn dialogues and medium-length document analysis.