Wanlian Securities: Smart Spectrum Releases AI Agent AutoGLM Meditation Will Usher in Smart Spectrum's “Open Source Year”

The Zhitong Finance App learned that Wanlian Securities released a research report saying that Zhi Spectrum's newly released AI agent product, Autoglm Meditation, is expected to promote technological progress and further upgrading of AI Agent applications. Compared to other agent products, AutoGLM Contemplation has three major characteristics: 1) the model is based on intelligent spectrum-based Z1 inference model to enhance learning training; 2) the model has GUI reading ability; 3) the model integrates the equipment operation capabilities of AutoGLM developed in-house by Intelligent Spectrum. At the same time, Smart Spectrum plans to officially open source its core link model and technology, which will help the AI Agent industry to build an overall ecosystem and accelerate the implementation of AI applications. It is recommended to focus on: 1) innovative development and accelerated implementation of AI agent products; 2) application upgrades of AI agents on the end side; 3) AI agent product acceleration application scenario implementation boosts computing power requirements.

The main views of Wanlian Securities are as follows:

Incident: On March 31, 2025, Zhipu officially released AutoGLM Meditation at the Zhongguancun Forum. According to reports, this is the first Agent integrating in-depth research and practical operation capabilities. Currently, Autoglm Meditation is online on the Zhipu Qingyan PC client, and users can experience its research and operation capabilities for free. This release is a preview version, with core support for research scenarios; the contemplation function has now been officially launched on the Zhipu Qingyan web, PC, and mobile apps, and is free and unlimited for everyone. This is also the first Deep Research feature to be officially opened in China.

AutogLM contemplation pushes AI agents into the “think while doing” stage. It has three key characteristics

AutoGLM contemplation is an autonomous agent (AI Agent) that can explore open questions and perform operations based on results. It not only has deep research capabilities (Deep Research), but also realizes actual operation (Operator), achieving the goal of “thinking while doing”. AutoGLM contemplation combines three major abilities: 1) deep thinking: being able to simulate human reasoning and decision-making processes in the face of complex problems; 2) sensing the world: being able to obtain and understand environmental information like a human; 3) tool use: being able to call and operate tools like humans to complete complex tasks.

Compared to other agent products, AutoGLM contemplation has three major characteristics: 1) The model is based on intelligent spectrum-based Z1 inference model, which is similar to Open AI's Deep Research training process based on the O3 model. The model can independently plan and make dynamic decisions according to different task goals, thereby continuously adjusting plans based on feedback without the need for a pre-designed workflow; 2) The model has GUI reading ability, and can not only call APIs, but also open and browse websites like humans. Browse dozens of web pages for tens of minutes to find all the information you want; 3) The model integrates the operating capabilities of AutoGLM's self-developed equipment. For example, after the report is completed, you can further send an email based on the results.

Smart Spectrum continues to lead AI Agent innovation and exploration, and will soon usher in Zhi Spectrum's “Open Source Year”

In the past, in AI Agent R&D exploration, Zhipu has always been in the leading echelon and continues to release innovative products. From the launch of Zhipu Qingyan and GLMS earlier, to the launch of the world's first device control agent, AutoGLM, in October 2024, to the previous three web-based, mobile (Android), and PC-oriented products based on AutoGLM upgrades on Agent OpenDay at the end of November 2024: AutoGLMWeb, AutoGLM (Android), and GLM-PC, SmartSpectrum has been leading the innovative exploration of AI agent products. The AutoGLM meditation launched this time can reflect Intellect Spectrum's core understanding of AI agents that “can not only think, but also take initiative”, and push AI agents to evolve from simple thinkers to intelligent executors that can deliver results.

The technical evolution path of AutoGLM contemplation includes: GLM-4 base model → GLM-Z1 inference model → GLM-Z1-rumination contemplation model → AutoGLM model. Among them, GLM-Z1-rumination breaks the limitations of traditional AI by simply relying on internal knowledge reasoning and innovatively combines real-time network search, dynamic tool call, in-depth analysis and self-verification to form a complete independent research process; the AutoGLM series model has also made significant progress once again, achieving SOTA results in 5 test environments on the intelligent benchmark evaluation AgentBench selected by the Stanford Model Center's “AI Index 2024”. Smart Spectrum plans to officially open source the core link model and technology on April 14 to promote the development of the industry ecosystem.

risk factors

Technological friction between China and the US; the competitiveness of large domestic AI models falls short of expectations; AI agent application implementation falls short of expectations; AI industry technology development and computing power requirements fall short of expectations.