智谱清言ChatGLM (Zhipu Qingyan)
智谱清言是「长文档解析 + 知识库检索」的国产对话产品：GLM 开源权重同族可选本地跑， 首批备案（2023-08-31）；产品名/模型名混用区，转述按官方页主语走。
Zhipu Qingyan (ChatGLM) is the domestic chat product built on long-document parsing plus knowledge-base retrieval: GLM open weights make local runs a same-family option; first-batch filed (2023-08-31); product/model names overlap — relay per the official pages' subjects.
适合与不适合When to use it, when not to
适合：中文长文档解析检索、GLM 权重云端+本地双选、文档密集场景。 不适合：要生态联动、要脱离账号。
Fits: Chinese long-document parsing and retrieval, GLM weights cloud+local, document-heavy work. Not a fit: ecosystem integration, account-free use.
中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-22
- 出品属地Origin
- 国产domestic 智谱 AI（北京智谱华章科技股份有限公司）。Zhipu AI (Beijing Zhipu Huazhang Technology Co., Ltd.).
- 国内使用Direct access
- 大陆可直连direct OK chatglm.cn 大陆直连实测 200（2026-10-22）。chatglm.cn measured 200 direct from mainland (2026-10-22).
- 备案状态CAC filing
- 已备案Filed：
Beijing-ChatGLM-20230821Beijing-ChatGLM-20230821公示名称「智谱清言(ChatGLM)」 listed as "Zhipu Qingyan (ChatGLM)" · 备案时间filed 2023-08-31 · 首批备案（备案时间 2023-08-31，属地北京）；备案号经阿里云百炼与火山方舟两页官方公示交叉核验。first-batch filing (2023-08-31, Beijing); filing no. cross-verified on Alibaba Cloud Model Studio and Volcengine Ark official disclosure pages.
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.
固定坐标系Fixed coordinate system
8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.
- 模型与开放条件Model & openness
主力是智谱自研 GLM 系列（智谱官网 zhipuai.cn 有模型发布线）；GLM 开源权重线（HF/ModelScope） 与云端对话同族——本地跑归 runner 档案。应用内模型档位以登录后界面为准。
The main line is Zhipu's self-developed GLM series (Zhipu's official site zhipuai.cn has a model-release line); the GLM open-weight line (HF/ModelScope) and the cloud chat are the same family — running locally belongs to the runner dossiers. In-app model tiers are per the signed-in UI.
- 运行位置Where it runs
网页（chatglm.cn）+ iOS/Android App；生成在厂商云端处理，本机不跑模型。 开发者入口是开放平台（open.bigmodel.cn），与对话产品同厂商两接口。
Web (chatglm.cn) + iOS/Android apps; generation is processed on the vendor's cloud, no model runs locally. The developer entry is the open platform (open.bigmodel.cn) — same vendor, two interfaces alongside the chat product.
- 本地文件Local files
对话历史与知识库文件存于账户侧；产品有「知识库」面（上传文档做检索问答），数据在云端账户。
Chat history and knowledge-base files are stored on the account side; the product has a "knowledge base" surface (upload documents for retrieval Q&A), data lives in the cloud account.
- 关机后的任务Tasks after shutdown
对话即时返回；产品含智能体与浏览器插件入口（官方导航所见），「关机续跑」形态未核验到官方声明。
Chats return instantly; the product includes agent and browser-extension entries (seen in official navigation); no official statement verified for an "offline continuation" form.
- 工具与扩展Tools & extensions
产品内含数据分析（代码解释器类）、文生图、联网检索、PPT 生成等能力线（官方功能页导航所见）； 智能体创建面存在。MCP 支持未逐条核验。
In-product capability lines incl. data analysis (code-interpreter type), text-to-image, web retrieval and PPT generation (seen in the official feature-page navigation); an agent-creation surface exists. MCP support not verified item by item.
- 上下文与记忆Context & memory
会话历史登录同步；知识库是显式的跨会话资料面；更细的记忆机制未逐条核验。
Session history syncs when signed in; the knowledge base is an explicit cross-session material surface; finer memory mechanics not verified item by item.
- 权限与限制Permissions & limits
账号体系（手机号注册）；训练用途开关路径未逐条核验。GLM 模型备案号见 cn 块（官方公示交叉核验）。
Account system (phone-number registration); the training-use switch path not verified item by item. The GLM model filing number is in the cn block (cross-verified against official public disclosures).
- 适合什么任务Best for
适合：中文长文档解析与检索问答、要 GLM 开源权重同族（云端不够就本地跑）、学生/职场文档场景。 不适合：要办公/社交生态联动、要完全脱离账号体系。
Suits: Chinese long-document parsing and retrieval Q&A, wanting the GLM open-weight same family (run locally if the cloud isn't enough), student/workplace document scenarios. Does not suit: office/social ecosystem integration, or fully leaving the account system.
头号误解Top misconception
- 把 chatglm.cn（对话产品）与 open.bigmodel.cn（开放平台 API）混为一谈——同厂商两接口，条款与配额各自成页Mixing chatglm.cn (chat product) with open.bigmodel.cn (open-platform API) — one vendor, two interfaces, separate terms and quotas
- 把 GLM 开源权重当产品本体——权重本地跑归 runner 生态，与云端对话的数据边界完全不同Treating GLM open weights as the product itself — running weights locally belongs to the runner ecosystem, with a completely different data boundary
- 「智谱清言」与「ChatGLM」与「GLM」是产品名/模型族名混用区——转述时主语按官方页走'Zhipu Qingyan' / 'ChatGLM' / 'GLM' is a name-overlap zone (product vs model family) — keep the subject per the official page when relaying
未知项清单Known unknowns
- 知识库容量/检索细节（未逐条核验）
- 训练用途开关路径与默认态（未逐条核验）
- 应用内模型档位分流（未登录核实）
- 智能体与浏览器插件能力边界（未逐条核验）
- Knowledge-base capacity/retrieval details (not verified item by item)
- The training-use switch path and default state (not verified item by item)
- In-app model tier routing (not verified while signed in)
- The capability boundary of agents and the browser extension (not verified item by item)
证据来源Evidence sources
判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-22，内容更新日content updated on 2026-10-22。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| official | 智谱清言 · 官网（产品与功能线）ChatGLM (Zhipu Qingyan) · official site (product and feature lines) | https://chatglm.cn |
| official | 智谱清言 · 服务条款（官方页，2026-10-22 直抓核验）ChatGLM (Zhipu Qingyan) · Terms of Service (official page, direct-fetch verified 2026-10-22) | https://chatglm.cn/terms |
| changelog | 智谱清言 · Blog（产品更新入口，changelog）ChatGLM (Zhipu Qingyan) · Blog (product-update entry, changelog) | https://chatglm.cn/blog |
| official | 阿里云百炼 · 模型备案信息公示（智谱清言(ChatGLM) 备案号交叉核验）Alibaba Cloud Model Studio · Model Filing Information Publicity (Zhipu Qingyan (ChatGLM) filing number cross-verification) | https://help.aliyun.com/zh/model-studio/model-filing-information-publicity |
实测记录Measured runs
本档未跑对话任务集；能力结论来自官方文档，不评价回答质量。
实测建议（三条，都指向官方可读页没给的东西）：
- 知识库容量上限与检索质量边界（官方有入口，上限未核验）
- 训练用途开关路径与默认态
- 应用内模型档位选择器实际形态
This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to chat products).
Suggested measurements (three, all pointing at things the official readable pages do not give):
- Knowledge-base capacity limits and retrieval-quality boundary (the official entry exists; limits unverified)
- The training-use switch path and default state
- The actual form of the in-app model tier selector
相关条目Related entries
- DeepSeekDeepSeek Chat同为国产开源权重同族：GLM 与 DeepSeek 权重都能在 runner 本地跑。Also a domestic open-weight same family: both GLM and DeepSeek weights can run locally under runner dossiers.
- KimiKimi同为国产长文档向：Kimi 主打超长上下文，清言主打知识库检索。Also domestic long-document-oriented: Kimi leads with ultra-long context, Qingyan leads with knowledge-base retrieval.
- OllamaOllama另一层 · 本地模型运行器Other layer · Local model runners
本页由This page is generated from the ai-handbook 数据层生成（CC BY 4.0）。data layer (CC BY 4.0).
方法论与坐标系定义见仓库内For the methodology and coordinate definitions, see METHODOLOGY.md.