跳到主要内容Skip to main content
← Chat

KimiKimi

月之暗面 Moonshot AIMoonshot AI
中文Chinese 长文档Long documents Agent任务Agent tasks 境内存储China-based storage partial lifecycle: active 核验Checked 2026-10-30

中文长文档对话的主力:K3/K2.7 Code/K2.6/K2.5 模型家族官方产品线清晰, 「退出训练要走客服登记、账号级统一」是最需要提前确认的一条流程。

The workhorse for long Chinese-document chat: the K3/K2.7 Code/K2.6/K2.5 model family is a clear official product line — “opting out of training goes through customer-service registration, at account level” is the process to confirm in advance.

适合与不适合When to use it, when not to

适合中文长文档解析、检索与 Agent 任务场景,境内存储对合规敏感场景是加分项。 在意「对话绝不进训练」要先接受客服登记流程(5-7 个工作日);企业数据走 Kimi Business。 要本机跑模型或自由换模型,它不合适——去看本地模型运行器与开源框架。

Chinese long-document parsing, retrieval and agent task scenarios; domestic storage is a plus for compliance-sensitive use. If “conversations never trained on” matters, first accept the customer-service registration flow (5–7 working days); enterprise data goes through Kimi Business.

Not a fit for running models 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
国产domestic 月之暗面(北京)。Moonshot AI (Beijing).
国内使用Direct access
大陆可直连direct OK kimi.com 大陆直连实测 200。kimi.com measured 200 direct.
备案状态CAC filing
已备案Filed:Beijing-MoonShot-20231016Beijing-MoonShot-20231016 公示名称「Moonshot(月之暗面)」 listed as "Moonshot (Moonshot AI)" · 备案时间filed 2023-11-03 · 首批公告第 16 行。Listed in the CAC registry as "Moonshot" (Moonshot AI; Beijing-MoonShot-20231016, filed 2023-11-03).

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

模型家族官方产品线披露:官网导航列 Kimi K3 / K2.7 Code / K2.6 / K2.5(2026-10-21 直抓核验), 首页标题「K3 上线,专为智能体编程与知识工作打造」;开放平台 API 另线提供模型接口。

The model family is disclosed via the official product line: the official-site navigation lists Kimi K3 / K2.7 Code / K2.6 / K2.5 (direct-fetch verified 2026-10-21), with the homepage headline "K3 is live, built for agentic coding and knowledge work"; the open-platform API provides model interfaces on a separate line.

运行位置Where it runs

网页为主入口,另有桌面端与移动应用;生成在云端处理,数据存储于中国境内服务器(官方数据使用与共享说明)。

Web is the main entry, plus desktop and mobile apps; generation is processed in the cloud and data is stored on servers within mainland China (official data usage & sharing notice).

本地文件Local files

长文档解析与检索是主打能力;支持 PPT/深度研究/文档/表格/构建应用等产品线入口(官网导航直抓核验); 上传内容存于账户。

Long-document parsing and retrieval are the flagship capabilities; product-line entries incl. PPT/deep research/documents/spreadsheets/build-apps are supported (verified via direct fetch of the official-site navigation); uploaded content is stored in the account.

关机后的任务Tasks after shutdown

官网导航有「定时任务」入口(2026-10-21 直抓核验);Agent 任务形态官方有产品线 (「尽管问,或做个 Agent 任务」);关机续跑细节未逐条核验。

The official-site navigation has a "scheduled tasks" entry (direct-fetch verified 2026-10-21); the Agent-task form has an official product line ("just ask, or make an Agent task"); offline-continuation details not verified item by item.

工具与扩展Tools & extensions

官方产品线含 PPT、深度研究、文档、表格、设计、构建应用与插件生态;另有开放平台 API(Moonshot 开放平台)。

The official product line includes PPT, deep research, documents, spreadsheets, design, build-apps and a plugin ecosystem; there is also the open-platform API (Moonshot open platform).

上下文与记忆Context & memory

有项目空间与历史会话组织(登录同步多端);跨会话记忆机制未逐条核验 ——官网有「灵感库」等入口,但独立记忆管理界面的官方文档本站未核验到。

Has project spaces and history-session organization (syncs across devices when signed in); the cross-session memory mechanism is not verified item by item — the official site has entries like an "inspiration library", but this site has not verified official docs for a standalone memory-management interface.

权限与限制Permissions & limits

官方数据使用与共享说明:在「经安全加密技术处理、严格去标识化且无法重新识别特定个人」前提下, 输入、指令与生成回复可能用于模型训练;退出须联系客服登记(验证身份后 5-7 个工作日), 账号级统一退出、不支持按对话退出,退出后历史与新增数据都不再进训练集。 Kimi Business 企业版承诺企业数据不进入模型训练流程,个人与企业空间完全隔离。

Official data usage & sharing notice: on the premise of "processing with secure encryption, strict de-identification such that specific individuals cannot be re-identified", inputs, prompts and generated replies may be used for model training; opting out requires registering with customer service (5-7 business days after identity verification), it is an account-level uniform opt-out with no per-chat opt-out, and after opting out neither historical nor new data enters the training set. Kimi Business promises enterprise data never enters the model training pipeline, with personal and enterprise spaces fully isolated.

适合什么任务Best for

中文长文档场景的主力。退出训练要走客服登记、不是自助开关——在意这一点要先确认; 企业数据走 Kimi Business。要本机跑模型,它不合适。

The workhorse for Chinese long-document scenarios. Opting out of training goes through customer-service registration, not a self-service switch — confirm this first if it matters to you; enterprise data goes through Kimi Business. For running models locally, it is not suitable.

头号误解Top misconception

  • 以为训练退出是自助开关——官方流程是客服登记、账号级统一退出Assuming the training opt-out is a self-serve toggle — the official flow is customer-service registration, an account-level unified opt-out
  • 把新旧两版协议混着读——旧版「仅非商业」与现行 v2「拥有输入输出 IP」并存,按现行版本走Mixing the old and new agreements — the old “non-commercial only” and current v2 “owns input and output IP” coexist; follow the current version
  • 把消费线与开放平台 API 混为一谈——两条线条款不同Confusing the consumer line with the open-platform API — the two lines have different terms

未知项清单Known unknowns

  • 跨会话记忆机制与定时任务的完整边界(官网有入口,本站未逐条核验)
  • 开放平台 API 与消费应用的训练条款差异细节(API 线条款本站未逐条核验)
  • The cross-session memory mechanism and the full boundary of scheduled tasks (the official site has entries; this site has not verified item by item)
  • Details of the training-terms difference between the open-platform API and the consumer app (the API line's terms not verified item by item 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 2026-10-30,内容更新日content updated on 2026-10-21。

类型Type名称Name链接Link
docsKimi · 数据使用与共享说明(官方帮助页,2026-10-20 直抓核验:训练前提、退出流程、境内存储、Business 不训练)Kimi · Data Usage & Sharing Notice (official help page, direct-fetch verified 2026-10-20: training premise, opt-out process, mainland storage, Business no-training)https://www.kimi.com/en/help/others/data-usage
docsKimi · 隐私政策(NOVASCENT PRIVATE LIMITED / Moonshot AI,2026-08-13 更新)Kimi · Privacy Policy (NOVASCENT PRIVATE LIMITED / Moonshot AI, updated 2026-08-13)https://www.kimi.ai/user/agreement/userPrivacy?version=v2
changelogKimi 官网(K3 上线标题与模型产品线导航,2026-10-21 直抓核验)Kimi official site (K3-live headline and model product-line navigation, direct-fetch verified 2026-10-21)https://www.kimi.com/
pricingKimi · 会员页(定价页指针,D9:本站不采价格数字)Kimi · membership page (pricing page pointer, D9: this site does not collect price numbers)https://www.kimi.com/membership

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).

相关条目Related entries

  • ChatGPTChatGPT通用对话对照:训练开关自助、反馈口径不同General-chat contrast: the training switch is self-service, feedback wording differs
  • 豆包Doubao同为国产通用对话:豆包 App 开关自助、Kimi 走客服登记Also domestic general chat: Doubao's App switch is self-service, Kimi goes through customer-service registration
  • Gemini(对话应用)Gemini (chat app)生态联动路线对照Ecosystem-integration route counterpart
  • xAI Grok BotxAI Grok Bot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.