AutoGLM
Zhipu AI · · Agent
Zhipu AI launched AutoGLM, an agent that operates a phone or a web browser on the user's behalf from a spoken or typed instruction. Its published architecture splits a Planner from a Grounder so that deciding what to do and locating the pixel to tap are separate models — the paper reports 55.2% success on web browsing and 36.2% on Android control.
Why it matters
China's answer to computer use, launched three months before OpenAI's Operator and two days after Anthropic's computer-use API. It aimed at the phone rather than the desktop, which is the split that has defined Chinese agent products since: Western agents drive a browser, AutoGLM drove WeChat, Taobao and Meituan.
This is LLM World's historical reading, not a claim by Zhipu AI. The facts above and the sources below are what it rests on.
Sources
Record status: partly verified · confidence 65/100
Dated by Yicai (第一财经), published 2024-10-30, which states 大模型独角兽公司智谱AI于10月25日正式上线AutoGLM新功能 ('the large-model unicorn Zhipu AI officially brought the AutoGLM feature online on 25 October'), with Wallstreetcn carrying the same launch. The AutoGLM paper — 'AutoGLM: Autonomous Foundation Agents for GUIs', submitted to arXiv 2024-10-28 — is a primary source for the system but was posted three days AFTER the launch and does not date it. Coverage also places a first demonstration at CNCC earlier in October 2024, which was not pinned down and is not claimed here. Zhipu's own dated announcement post could not be read. NOT A CODING AGENT: AutoGLM is a GUI-control agent, included because it is Zhipu's agent-era starting point and the Chinese counterpart to Operator and computer use.