ChatGPTChatGPT
English summaries and interface labels are available. Detailed source notes below are currently in Chinese.
OpenAI 的通用对话助手:网页/应用/桌面全端覆盖,模型家族按档位分流, 数据训练开关是它最常被误读的一行条款。
OpenAI’s general-purpose chat assistant: web, apps and desktop fully covered, model family split by subscription tier — and the data-training toggle is its most commonly misread clause.
适合与不适合When to use it, when not to
适合什么都能问一点的通用问答、写作与文件分析,前提是接受云端处理。 在意「对话绝不进训练」就关数据控制开关或用临时聊天;企业数据走 Team/Enterprise/API。 要本机运行或自由换模型,它不合适——去看本地模型运行器与开源框架。
General Q&A, writing and file analysis where you accept cloud processing. Users who want “conversations never trained on” should turn off the data-improvement toggle or use Temporary Chat; enterprise data goes through the Team/Enterprise/API line.
Not a fit for running locally or freely swapping models — see the local model runners and open-source frameworks.
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 2026-10-20
- 出品属地Origin
- 国外出品foreign-made 国外主体(网信办公示名单未见对应备案主体)。foreign entity (no matching filing entity in the CAC registry).
- 国内使用Direct access
- 直连不可达direct blocked chatgpt.com 大陆直连稳定超时(2026-10-20 同日两次 000)。chatgpt.com times out consistently from mainland (000 twice).
- 备案状态CAC filing
- 未见于备案名单not in filing list:未见于网信办公示名单(2026-10-20 检索 13 批附件,https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm);该产品未面向大陆公众提供服务,备案不适用。not in the CAC public registry (13 batch attachments searched 2026-10-20, https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm); the product does not serve the mainland public, so filing does not apply.
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
模型家族官方披露并按档位分流:release notes(2026-10-07)写明 Plus/Pro/Business/Enterprise 用 GPT-6 Sol、Free/Go 用 GPT-6 Luna,Pro 推理选项用 GPT-6 Astra;聊天内可选模型。
The model family is officially disclosed and routed by tier: release notes (2026-10-07) state Plus/Pro/Business/Enterprise use GPT-6 Sol, Free/Go use GPT-6 Luna, and Pro reasoning options use GPT-6 Astra; models are selectable in chat.
- 运行位置Where it runs
网页、iOS/Android、macOS/Windows 桌面端;全部生成与检索在 OpenAI 云端处理,本机不跑模型。 桌面端内置浏览器,浏览器智能体能力并入 ChatGPT 与 Codex(官方版本说明:Atlas 浏览器于 2026-08-09 停运)。 桌面端快捷聊天/截屏功能也只是入口,处理仍在云端。
Web, iOS/Android, macOS/Windows desktop apps; all generation and retrieval are processed on OpenAI's cloud — no model runs locally. Desktop quick-chat/screenshot features are just entry points; processing still happens in the cloud.
- 本地文件Local files
可上传文件、图像与音频(音频上传见 release notes 2026-10-06:转写、摘要、问答); 上传内容存于 OpenAI 账户,归谁与留存见 permissions 维。
Can upload files, images and audio (audio upload per release notes 2026-10-06: transcription, summarization, Q&A); uploaded content is stored in the OpenAI account — ownership and retention are covered by the permissions dimension.
- 关机后的任务Tasks after shutdown
对话即时返回。ChatGPT 工作(官方版本说明,2026-07)面向长任务:可跟踪进度、回答问题、调整方向、 批准重要操作,并支持一次性或按计划/触发条件重复运行。Codex Cloud 线明确「任务可在你电脑休眠时继续」 (release notes 2026-09-29)。对话产品本体没有关机续跑形态。
Chats return instantly; the Codex Cloud line explicitly says "tasks can continue while your computer sleeps" (release notes 2026-09-29). The chat product itself has no offline-continuation form.
- 工具与扩展Tools & extensions
Deep Research(官方 2026-02-10 更新:可连接任意 MCP 或 app、把网页搜索限定到可信站点、实时进度并可中断补充); ChatGPT 智能体 agent mode(公开网站、上传文件、已连接第三方来源,官方 2025-07-17 发布、2025-08-08 向 Enterprise/Edu 推出); ChatGPT 图像 2.0(2026-04-21 上线;官方 DALL·E GPT 于 2026-08-30 停用); 官方连接器(Gmail/Calendar 等)与合作 Apps;Slides/Sites/Finances 等内建功能线;另有独立 API 线。
Official connectors (Gmail/Calendar etc.) and partner Apps; built-in feature lines like Slides/Sites/Finances; a separate API line exists. Plugins are disclosed on the Codex line; the consumer chat surface extends mainly via official connectors.
- 上下文与记忆Context & memory
有记忆功能与 Projects 组织历史;临时聊天不进历史、不建记忆。 跨会话记忆的完整边界以官方帮助页为准。
Has memory and Projects for organizing history; temporary chats leave no history and build no memory. The full boundary of cross-session memory is per the official help pages.
- 权限与限制Permissions & limits
官方政策页:个人服务默认可能用对话训练模型,可在「设置 → 数据控制」关闭或在隐私门户选 「请勿使用我的内容训练」;临时聊天不训练;Team/Enterprise/API 默认不训练(API 需显式 opt-in); 你提交的点赞/点踩反馈,其整段对话仍可能用于训练。
Official policy pages: personal services may use chats for model training by default; you can turn this off under "Settings → Data controls" or choose "Don't use my content for training" in the privacy portal; temporary chats are not used for training; Team/Enterprise/API do not train by default (API requires explicit opt-in); for the thumbs up/down feedback you submit, the whole related conversation may still be used for training.
- 适合什么任务Best for
适合「什么都能问一点」、接受云端处理的通用问答与写作。敏感内容先关训练开关或用临时聊天; 企业数据走 Team/Enterprise/API 线。要本机处理或可换模型,这类闭源云助手都不合适。
Suits "ask anything a bit" general Q&A and writing, provided you accept cloud processing. For sensitive content, turn off the training switch first or use temporary chat; enterprise data goes through Team/Enterprise/API. For local processing or swappable models, closed cloud assistants like this are not suitable.
头号误解Top misconception
- 把「关闭训练开关」当成临时聊天——两套独立机制,开关只管未来对话Treating "turn off training" as temporary chat — two separate mechanisms; the switch only affects future chats
- 以为反馈(点赞/点踩)不受训练开关约束——官方明说整段相关对话可能入训练Assuming feedback (thumbs up/down) is exempt from the training switch — the official docs say the whole related conversation may be used
- 把消费线条款套到 API/Enterprise——两条线数据条款不同Applying consumer terms to API/Enterprise — the two lines have different data terms
- 还在找 ChatGPT Atlas 浏览器——官方已宣布 2026-08-09 停运,浏览器智能体能力并入 ChatGPT 与 CodexStill looking for ChatGPT Atlas — the official notice says it was retired on 2026-08-09 and browser-agent capability moved into ChatGPT and Codex
- 以为 DALL·E GPT 仍是现役图像入口——官方于 2026-08-30 停用,改用 ChatGPT 图像Assuming DALL·E GPT is still the live image entry — the official notice retired it on 2026-08-30; use ChatGPT Images
未知项清单Known unknowns
- 跨会话记忆功能的完整边界(以官方帮助页为准,本站未逐条核验)
- The full boundary of the cross-session memory feature (per the official help pages; not item-by-item verified by this site)
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 2026-11-03,内容更新日content updated on 2026-11-03。
| 类型Type | 名称Name | 链接Link |
|---|---|---|
| changelog | OpenAI Help · ChatGPT release notes(官方更新日志;2026-11-03 取到正文,含 Atlas 停运、ChatGPT 工作、DALL·E GPT 停用)OpenAI Help · ChatGPT release notes (official changelog; body captured 2026-11-03, incl. Atlas retirement, ChatGPT Work and DALL·E GPT retirement) | https://help.openai.com/zh-hans-cn/articles/6825453-chatgpt-%E7%89%88%E6%9C%AC%E8%AF%B4%E6%98%8E |
| changelog | OpenAI Help · ChatGPT Atlas release notes(Atlas 版本记录,2026-03-10 最后一条)OpenAI Help · ChatGPT Atlas release notes (Atlas version record, last entry 2026-03-10) | https://help.openai.com/zh-hans-cn/articles/12591856-chatgpt-atlas-release-notes |
| changelog | OpenAI Help · ChatGPT 智能体 release notes(2025-07-17 发布 / 2025-08-08 Enterprise+Edu)OpenAI Help · ChatGPT agent release notes (launched 2025-07-17; Enterprise+Edu 2025-08-08) | https://help.openai.com/zh-hans-cn/articles/11794368-chatgpt-agent-release-notes |
| docs | OpenAI · Introducing deep research(2025-02-02 发布;2026-02-10 更新 MCP/app 连接与可信站点限制)OpenAI · Introducing deep research (released 2025-02-02; 2026-02-10 update added MCP/app connections and trusted-site limits) | https://openai.com/index/introducing-deep-research/ |
| docs | OpenAI · ChatGPT 图像 2.0 现已上线(2026-04-21)OpenAI · ChatGPT Images 2.0 is here (2026-04-21) | https://openai.com/zh-Hans-CN/index/introducing-chatgpt-images-2-0/ |
| docs | OpenAI · How your data is used to improve model performance(官方政策页,2026-10-20 快照核验)OpenAI · How your data is used to improve model performance (official policy page, snapshot-verified 2026-10-20) | https://openai.com/policies/how-your-data-is-used-to-improve-model-performance |
| docs | OpenAI · Terms of Use(内容所有权与转让条款,2026-10-21 经搜索快照核验)OpenAI · Terms of Use (content-ownership and assignment clauses, verified via search snapshot 2026-10-21) | https://openai.com/policies/row-terms-of-use/ |
| pricing | OpenAI · ChatGPT Pricing(定价页指针,D9:本站不采价格数字)OpenAI · ChatGPT Pricing (pricing page pointer, D9: this site does not collect price numbers) | https://openai.com/chatgpt/pricing/ |
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。
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).
第一次怎么用
1. 登录后先开设置里的数据开关:把「用于改进服务」与「训练」分别看清楚,再决定是否关闭。 2. 上传一个真实文件(PDF 或表格),要求它先给目录与结论,再逐段引用原文。 3. 需要多步研究时,明确写「先列检索计划,再执行,最后给带日期的来源表」。 4. 要图像就直说尺寸、风格与用途;图像 2.0 已并入主产品,不需要另找入口。 5. 需要浏览器操作时注意:Atlas 已于 2026-08-09 停运,能力并入 ChatGPT / Codex,入口变了。
请基于我上传的文件回答,规则:
1) 先列 3 条结论,每条后面标原文页码;
2) 再列文件里没有覆盖的问题,写「未找到」;
3. 不要引用文件之外的信息。
相关条目Related entries
- Claude(对话应用)Claude (chat app)同为通用对话,训练 opt-out 机制与「安全审查例外」不同Also general chat; the training opt-out mechanism and the "safety-review exception" differ
- Gemini(对话应用)Gemini (chat app)生态联动路线:Google 系用户的对照选择Ecosystem-integration route: the counterpart choice for Google-family users
- 豆包Doubao中文场景对照:App 训练开关自助、API 默认不训练Chinese-context counterpart: App training switch is self-service, API does not train by default
- OpenAI DotOpenAI Dot另一层 · AI 编程Other layer · AI coding
- Codex CLICodex CLI另一层 · AI 编程Other layer · AI coding
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.