# ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools

**Authors:** Team GLM,  :, Aohan Zeng, Bin Xu, Bowen Wang, Chenhui Zhang, Da Yin, Dan Zhang, Diego Rojas, Guanyu Feng, Hanlin Zhao, Hanyu Lai, Hao Yu, Hongning Wang, Jiadai Sun, Jiajie Zhang, Jiale Cheng, Jiayi Gui, Jie Tang, Jing Zhang  
**arXiv:** [2406.12793](https://arxiv.org/abs/2406.12793)  
**Date:** 2024-06-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2406.12793v2/

## Key points

- Ranging from GLM-130B to GLM-4 (All Tools)
- Significant advancements made by the team in understanding and developing these models over the past year and a half
- GLM-4 (0116, 0520), GLM-4-Air (0605), and GLM-4 All Tools
- Autonomously utilizing external tools and functions for handling complex tasks effectively
- Performance on par with or surpassing state-of-the-art models like GPT-4 Turbo, Claude 3 Opus, Gemini 1.5 Pro
- Particularly excelling in tasks related to the Chinese language
- Commitment to promoting accessibility and safety in Large Language Models (LLMs):
- Open release of model weights and techniques developed throughout the journey
- Over 10 million downloads on platforms like Hugging Face in 2023 alone
- Future focus on democratizing cutting-edge LLM technologies through open sourcing:
- Developing even more capable models based on gained knowledge
- Pushing boundaries of model capabilities towards teaching machines to think like humans
- Gratitude extended to contributors including data annotators, infrastructure operating staff members, collaborators, partners at Zhipu AI, Tsinghua University, Yuxuan Zhang, Wei Jia from Zhipu AI, teams at Hugging Face, ModelScope, WiseModel for their support in open-sourcing efforts

## Summary

Introducing the ChatGLM family of large language models: from GLM-130B to GLM-4 (All Tools). Our team has made significant advancements in understanding and developing these models over the past year and a half. With each generation, we have implemented more effective strategies for model pre-training and alignment. The latest ChatGLM models - including GLM-4 (0116, 0520), GLM-4-Air (0605), and GLM-4 All Tools - showcase remarkable progress in handling complex tasks by autonomously utilizing external tools and functions. These GLM-4 models have demonstrated performance on par with or even surpassing state-of-the-art models like GPT-4 Turbo, Claude 3 Opus, and Gemini 1.5 Pro - particularly excelling in tasks related to the Chinese language. Our commitment to promoting accessibility and safety in Large Language Models (LLMs) is evident through the open release of model weights and techniques developed throughout our journey. The open-sourcing efforts of our models have been well-received, with over 10 million downloads on platforms like Hugging Face in 2023 alone. We are currently working on developing even more capable models based on the knowledge gained so far. Moving forward, our focus remains on democratizing cutting-edge LLM technologies through open sourcing while pushing the boundaries of model capabilities towards the goal of teaching machines to think like humans. We extend our gratitude to all data annotators, infrastructure operating staff members, collaborators, partners at Zhipu AI and Tsinghua University who have contributed to the success of ChatGLM. Special thanks to Yuxuan Zhang and Wei Jia from Zhipu AI as well as teams at Hugging Face, ModelScope, WiseModel for their support in open-sourcing efforts.