Meet Tsinghua University's GLM-4-9B-Chat-1M: An Outstanding Language Model Challenging GPT 4V, Gemini Pro (on vision), Mistral and Llama 3 8B

GLM-4 9B's architecture differs from other industry models like GPT and Gemini in a few key ways. Firstly, it supports high-resolution vision tasks up to 1198 x 1198 pixels, which is a unique feature not commonly found in other language models. Additionally, GLM-4 9B has a diverse range of language capabilities, as it is trained on an unprecedented 10 trillion tokens spanning 26 languages4. This multilingual proficiency sets it apart from other models that may focus primarily on English or a smaller set of languages.

Furthermore, GLM-4 9B's architecture is built upon the latest advancements in deep learning, incorporating cutting-edge techniques such as attention mechanisms and transformer architectures6. This allows for better context understanding and more accurate predictions. The base version supports a context window of up to 128,000 tokens, while a specialized variant allows for an impressive 1 million token context length, outperforming many other models in this regard.

The article does not provide specific benchmarks in which GLM-4 9B outperformed GPT-4 and Gemini. However, it mentions that GLM-4 9B surpassed every other existing model in overall accuracy and outperformed GPT-4, Gemini Pro (in vision tasks), Mistral, and Llama 3 8B.