Zhipu AI has announced the launch of the GLM-4-9B series models, including base models, Chat models with varying context lengths, and visual models, all of which have surpassed the capabilities of the LLaMA38B. It is reported that the GLM-4-9B series models have been open-sourced on Github, attracting significant attention from developers and researchers. The release of this series of models is considered another significant breakthrough in the field of artificial intelligence by Zhipu AI.
GLM-4-9B is the latest open-source version of the GLM-4 series, Zhipu AI's newest generation of pre-trained models. In evaluations across datasets in semantics, mathematics, reasoning, coding, and knowledge, both GLM-4-9B and its human-aligned version, GLM-4-9B-Chat, have demonstrated superior performance over Llama-3-8B.
In addition to multi-round dialogue capabilities, GLM-4-9B-Chat also features advanced functionalities such as web browsing, code execution, custom tool invocation (Function Call), and long text reasoning (supporting up to 128K context).
This generation of models has added multilingual support, including 26 languages such as Japanese, Korean, and German. We have also introduced the GLM-4-9B-Chat-1M model, which supports a 1M context length (approximately 2 million Chinese characters), and the multimodal model GLM-4V-9B based on GLM-4-9B.
GLM-4V-9B possesses the capability for Chinese-English multi-round dialogues at a high resolution of 1120*1120. In comprehensive evaluations of Chinese-English abilities, perceptual reasoning, text recognition, and chart understanding, GLM-4V-9B has demonstrated superior performance over GPT-4-turbo-2024-04-09, Gemini1.0Pro, Qwen-VL-Max, and Claude3Opus.
Netizens have expressed hope that the GLM-4-9B series models will bring new momentum to the development of artificial intelligence technology and look forward to seeing more intelligent products come to market. They have also expressed admiration for Zhipu AI's technical prowess and innovative capabilities.
Open-source address: https://github.com/THUDM/GLM-4
Model experience address: https://modelscope.cn/studios/dash-infer/GLM-4-Chat-DashInfer-Demo/summary