Zhipu AI Releases Open-Source Version of GLM-4-9B 175

Zhipu AI launches the open-source version GLM-4-9B of the latest generation pre-trained model GLM-4 series, including the base model, Chat models with different context lengths, and a vision model. Faster speed, lower price, 40% performance improvement, comprehensively surpassing LLaMA 3 8B. Comparable to GPT-4V, supporting 1M lossless context, 26 languages, with a 40% improvement in function calling capability, on par with GPT-4V.

GLM-4-9B is the open-source version of the latest generation pre-trained model GLM-4 series launched by Zhipu AI. In evaluations on datasets covering semantics, mathematics, reasoning, code, and knowledge, GLM-4-9B and its human preference-aligned version GLM-4-9B-Chat both demonstrate outstanding performance surpassing Llama-3-8B. In addition to multi-turn dialogue, GLM-4-9B-Chat also features advanced capabilities such as web browsing, code execution, custom tool calling (Function Call), and long-text reasoning (supporting a maximum context of 128K). This generation of models adds multilingual support, covering 26 languages including Japanese, Korean, and German. We have also launched the GLM-4-9B-Chat-1M model that supports a context length of 1M (approximately 2 million Chinese characters), as well as the multimodal model GLM-4V-9B based on GLM-4-9B. GLM-4V-9B features bilingual (Chinese and English) multi-turn dialogue capability at a high resolution of 1120 * 1120. Across multiple multimodal benchmarks, including comprehensive Chinese and English capabilities, perception and reasoning, text recognition, and chart understanding, GLM-4V-9B demonstrates outstanding performance surpassing GPT-4-turbo-2024-04-09, Gemini 1.0 Pro, Qwen-VL-Max, and Claude 3 Opus.

GLM-4-9B Github Address:https://github.com/THUDM/GLM-4