Zhispu Qingyan

What is Zhipu Qingyan?

Zhipu Qingyan is a Chinese AI assistant and Agent platform developed by Beijing Zhipu Huazhang Technology Co., Ltd. It leverages the GLM large model developed by Zhipu to offer functions such as question answering, searching, writing, learning, programming, document processing, PPT creation, data analysis, and the execution of multi-step tasks. Initially it was known for its ChatGLM dialogue feature, but its role has now evolved from simply providing answers to understanding objectives, planning steps, utilizing various tools, and delivering finished results.

As of this verification, the model currently displayed on the Zhipu Qingyan official website isGLM-5.2This model is designed for long-term tasks and Agentic Engineering; it supports a context size of up to 1 million tokens and output sizes of up to 128K, while also offering different levels of reasoning strength.

For end-users, there are also options such as research reports, PPT creation, data analysis, AgentMore for multi-agent group chats, and GLM-Claw. Older articles often referred to GLM-4, GLM-4-Plus, or GLM-Z1 as the latest models, but these are no longer the current versions.

Core functions

• General Q&A and in-depth thinking:Complete tasks involving knowledge quizzes in Chinese and English, explanations, translation, writing, as well as mathematical and logical analysis. Simple tasks can be answered directly, while for complex problems the thinking mode can be used; a longer reasoning process does not guarantee that the conclusion will be correct.
• Agent task execution:Based on the automated step-by-step breakdown of objectives, it utilizes tools for searching, data processing, plotting, coding, and file management, followed by self-checking and summarization. It is suitable for creating travel itineraries, planning events, conducting industry research, and carrying out multi-phase office tasks.
• Research Report:It involves conducting online searches related to a specific theme, organizing information into an outline, analyzing the data available, and producing structured reports; it is suitable for preliminary research in industries, brands, finance, and academia. Citations, financial figures, and real-time events still need to be verified against the original sources.
• PPT creation:After entering the topic or material, it generates a presentation structure, page text, and templated visuals, which can be further modified and exported. It reduces the time required to create drafts, but data charts, brand guidelines, fonts, and copyrighted materials need to be reviewed manually.
• Data analysis:Read tables or data from multiple sources, carry out cleaning, feature preparation, statistical analysis, as well as the creation of charts and the presentation of conclusions. It is possible that the wrong parameters, formulas, or modeling methods are chosen for the model; therefore, important business conclusions cannot be drawn solely based on the generated reports.
• Websites and programming:It allows for writing and running code, as well as generating H5 or modular web pages; it also helps with debugging, testing, and project restructuring. GLM-5.2 places particular emphasis on the project-level engineering context and long-term development, enabling the progression from initial requirements to deployable products across multiple platforms.
• File processing:Read, summarize, and organize documents; extract fields, compare contents, and generate new files. Scanned documents, complex tables, documents with a mix of text and images, and large files may still not be recognized completely.
• Advanced search:It enables in-depth online searches for up-to-date information and the synthesis of answers, making it suitable for researching trending topics, products, policy backgrounds, and competitors. The search results come from third-party websites, and the timeliness and accuracy of these results are not guaranteed by the model.
• Learning and academic support:It offers assistance with study plans, explanation of topics, essay frameworks, organization of literature, data analysis, and writing standards. This can help improve the efficiency of research, but it is not permissible to write essays on behalf of others, fabricate experiments, or invent citations.
• Sound and multimedia creation:The mobile version offers features such as the ability to find study partners, select different sound tones, and clone voices; it also provides tools for drawing and video creation. Different generation capabilities may be available depending on points earned, membership level, or eligibility for specific campaigns.

AgentMore and GLM-Claw

AgentMoreIt is Zhipu Qingyan’s multi-agent collaboration capability. Users can, with just one sentence, recruit multiple Agents of different professions, personalities, or perspectives; they can decide who will participate and how the tasks will be divided, and they can switch between allowing free speech and restricting speeches to specific situations.

It is suitable for brainstorming, debate of opinions, cross-role reviews, and workflows that range from data collection to the creation of written content. Multiple agents increase the number of calls and the amount of context used, and they can also lead to situations where agents agree with each other too much, responsibilities become unclear, or errors are repeated.

GLM-ClawIt is designed as a sustainable, evolving AI entity that can work in conjunction with points and long-term tasks. It places greater emphasis on continuous operation and autonomous execution than on one-time conversations, which is why restrictions are imposed on accounts, files, external tools, and the ability to post automatically.

Stock trading, payments, sending messages, modifying data, or making public announcements should not be carried out without supervision.

Capabilities of the GLM-5.2 model

GLM-5.2 is the latest flagship text model recommended by the Zhipu Open Platform; it supports a context volume of up to 1 million tokens, a maximum output of 128K tokens, streaming responses, function calling, context caching, JSON-formatted output, and MCP tool calls.

It allows for the setting of a thinking switch, and enables selection between different levels of reasoning intensity via the \"reasoning effort\" parameter; it is suitable for tasks such as understanding large-scale warehouses, restructuring data across files, reproducing scientific research results, developing mobile applications, creating mini-programs, and executing long-running tool commands.

The GLM-5.2 version with publicly available weights has 744 billion total parameters and 40 billion activation parameters; it is available in BF16 and FP8 formats. A context size of around 1 million is suitable for project-level data, but actual usage is still limited by factors such as the model’s interface, GPU memory, costs, and task duration.

The fact that consumers can use GLM-5.2 does not mean that every account can generate 1 million tokens without any restrictions.

Price and version comparison

Free use and member prices

Zhipu Qingyan allows for free registration, enabling users to experience basic Q&A functions as well as some Agent features. Advanced models, higher usage volumes, GLM-Claw, generation tasks, and other additional benefits require a membership status, SVIP tier, or points. The free quota, number of uses per day, and benefits offered through promotions change over time; therefore, the claim of \"unlimited free use\" from previous promotions should not be regarded as a permanent guarantee.

The main in-app purchases currently listed on the App Store in China include:

• Monthly subscription: 79 yuan;
• Monthly subscription with consecutive billing: 219 yuan;
• Continuous annual subscription: 859 yuan;
• Qingyan Monthly Card member: 119 yuan;
• Qingyan SVIP monthly subscription or SVIP monthly card: 229 yuan;
• Subscription packages: 29 yuan, 109 yuan, 699 yuan, and other options.

In the app description, the old promotional figures such as 19 yuan for a monthly subscription, 79 yuan for a quarterly subscription, and 299 yuan for an annual subscription can still be seen, but the current in-app purchase list shows a different set of prices. When there is a discrepancy between these two sets of information, it is the purchase confirmation pop-up and Apple’s confirmation page that determine the actual amount to be charged.

Different membership levels may include varying numbers of models, usage limits, points, Agents, and generation privileges; it is not possible to judge them solely based on price or name.

A recurring subscription is usually renewed automatically 24 hours before it expires; to cancel it, one must go to the subscription management section in the Apple account. As for whether purchased credits expire, can be carried over, or if failed transactions are refunded, the rules applicable to the account shall govern these matters.

API prices and developers’ capabilities

The Zhipu AI Open Platform and the Zhipu Qingyan membership are separate products. Developers need to register for the open platform separately, create an API Key, and pay on a usage-based basis; Qingyan membership points cannot be used to offset API costs.

Currently, GLM-5.2 is priced per million tokens: 8 yuan for input, 28 yuan for output, and 2 yuan for cache hits; cache storage is available at no cost for a limited time.

The same price page also offers GLM-5, GLM-5-Turbo, GLM-4.7, visual understanding, and free Flash models. When cost is a priority, one can choose the smaller or free models.

Use GLM-5.2 only when around 1 million contexts and complex, long-term tasks are required.

Caching, search, knowledge base, image, video, and voice capabilities may incur additional charges; it is necessary to set budgets and alerts in the control panel before integrating them officially.

The open platform offers native SDKs as well as OpenAI-compatible interfaces that enable dialogue, streaming output, tool invocation, and structured results. Compatibility here refers to similar request formats; it does not guarantee that the Responses API, thought chains, tool streaming parameters, and the behavior of all third-party frameworks will be exactly identical. When migrating Agent projects, it is necessary to conduct multiple rounds of actual testing.

Open-source status and local deployment

The weights of the GLM-5.2 model have been made public by the developers, and Hugging Face model cards use them explicitly.MIT LicenseThere are no geographical restrictions; the official GitHub repository is licensed under Apache 2.0, and it provides explanations of the models, examples, as well as information on how to deploy them.

The weight volume for BF16 is approximately 1.51 TB; the developers have also released an FP8 version, which supports inference frameworks such as SGLang, vLLM, Transformers, KTransformers, Unsloth, and those related to Ascend.

Therefore, ZhiPu QingYan cannot be simply classified as either “fully open source” or “fully closed source”. A more accurate description is that the weights of core models such as GLM-5.2 and some of the code are available under an open source license, but the ZhiPu QingYan website, its app, AgentMore, GLM-Claw, as well as the search functionality, account systems, membership features, and cloud-based workflows are not made available as part of an open source product.

Even if the model is licensed under the MIT license, local deployment still entails responsibilities regarding hardware, data, compliance with regulations, and brand usage.

Supported platforms

Zhipu Qingyan offers access via web interfaces as well as clients for iPhone, iPad, and Mac. For the Apple apps available in China, iOS or iPadOS 14.0 or later is required, while macOS 10.13 or later is needed for Mac devices; Apple Vision also supports corresponding versions.

New features for mobile and web versions are usually released in batches; there may be version differences regarding AgentMore, file export, or certain generation capabilities.

Guide to Using Zhipu Qingyan

Complete a basic task.

1. Specify the presentation topic, objectives, duration, and number of pages;
2. Import reliable data into Zhipu Qingyan and remove any irrelevant or sensitive information;
3. Use general Q&A and deep thinking to generate an outline rather than arriving at a final version directly;
4. Select templates, diagrams, and visual hierarchies through Agent task execution;
5. Check each page for numbers, citations, charts, font sizes, and layout;
6. Perform a full-screen preview before exporting in the desired format;

Create reusable professional workflows

1. Create brand templates, title hierarchies, and a library of common pages;
2. Fix the order of general Q&A and in-depth thinking, Agent task execution, and report preparation;
3. Specify one key point per page and control the text density;
4. Assign the review of facts, visuals, and presentation logic to respective responsible persons;
5. Test fonts, animations, and proportions on different devices;
6. Save the source files, versions, and records of final approval;

Which users are it suitable for

• Individual users who need assistance with Chinese question-and-answer, searching, writing, translation, and learning;
• Professionals who need research reports, PPTs, data analysis in table form, and industry research;
• Creators who wish to use multiple agents for brainstorming, role evaluation, and collaborative workflow management;
• Developers who need long-distance code agents, an understanding of project-level repositories, and the ability to invoke tools;
• Companies and research teams that wish to develop private AI applications using APIs or open-source weights.

Usage restrictions and safety precautions

• AI-generated facts, paper citations, financial figures, code, and legal provisions can all be incorrect;
• Research reports and advanced searches should retain traceable information, the charts in PPTs must correspond to the original data, codes need to be tested and scanned for security, and medical or investment-related content cannot replace professional advice.
• Before uploading papers, contracts, financial reports, customer forms, source code, and audio samples, personal identifiers, keys, and trade secrets should be removed.
• According to the official privacy policy, a mobile phone number is required for account registration, and the content submitted by users is processed based on questions and answers, as well as various specific functions.
• Voice cloning and character assets require the permission of the respective owners, and must not be used for impersonation, fraud, or deception.
• The capabilities of agents can amplify the risks associated with ordinary conversations.
• Configure minimum permissions for GLM-Claw, AgentMore, and web development tasks, restricting the accessible directories, external accounts, and network services.
• Manual approval is required when it comes to deletion, sending, payment, deployment, and public release.

Frequently Asked Questions

Is Zhipu Qingyan free?

The basic functions can be tried out for free, but advanced models, more Agent quotas, SVIP status, and generation tasks may require a membership or points. The current monthly membership fee is 79 yuan; the actual price is subject to what is displayed on the purchase page.

What model is Zhipu Qingyan using now?

The official website currently displays GLM-5.2. It supports a context size of 1 million tokens, an output capacity of up to 128K, as well as deep reasoning and tool invocation; it is the latest flagship text model available.

What is AgentMore?

It is a multi-agent group chat and collaboration feature; users can define the occupations, personalities, and roles of multiple agents, allowing them to discuss freely or speak as needed in order to accomplish complex tasks together.

Can Zhipu Qingyan be used for creating PPTs and conducting data analysis?

Yes. The product offers tools for creating PPTs, preparing research reports, and conducting data analysis; it allows for the retrieval of information, the generation of pages and charts, and the derivation of conclusions. However, the final content still requires manual review.

Does the Zhipu Qingyan membership include an API?

It is not included. The membership levels for consumers on the Qingyan platform, as well as SVIP status and points, are billed separately from the APIs of the Zhipu Open Platform; for those APIs, separate keys must be created and top-ups must be made.

Is GLM-5.2 open source?

The official sources have made the weights for BF16 and FP8 available, and the model is licensed under the MIT license; however, the complete products of Zhipu Qingyan, its Agent services, and its cloud platform are not fully open source.