← Chat

腾讯元宝Tencent Yuanbao

腾讯Tencent

国产China-origin 免费+登录Free + sign-in 微信生态WeChat ecosystem 公众号信源WeChat public-account sources 文档精读Document analysis 已备案Filed partial lifecycle: active 核验Checked 2026-10-22

腾讯元宝是泡在微信生态里的通用 AI 助手：搜公众号/视频号、评论区 @ 解读、联动腾讯文档与微信读书，长文档精读是强项；模型主力是腾讯自研混元（Hy 系列）。

Tencent Yuanbao is the general AI assistant living in the WeChat ecosystem: search official accounts/Channels, @-mention it in comment sections, integrate Tencent Docs and WeChat Read; long-document reading is its strong suit, powered by Tencent's self-developed Hunyuan (Hy) models.

适合与不适合When to use it, when not to

适合：微信生态用户、长文档精读、免费额度。不适合：要本机跑模型、要脱离腾讯账号。

Fits: WeChat-ecosystem users, long-document reading, free allowance. Not a fit: running models locally, leaving the Tencent account system.

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　腾讯（深圳市腾讯计算机系统有限公司）。Tencent (Shenzhen Tencent Computer Systems Company Limited).
国内使用Direct access: 大陆可直连direct OK　yuanbao.tencent.com 大陆直连实测 200（2026-10-22）。yuanbao.tencent.com measured 200 direct from mainland (2026-10-22).
备案状态CAC filing: 已备案Filed：Guangdong-TencentHunyuan-20230901Guangdong-TencentHunyuan-20230901　公示名称「腾讯混元助手大模型（元宝所用模型）」　listed as "Tencent Hunyuan Assistant LLM (the model Yuanbao uses)"　·　元宝所用混元模型备案号经腾讯云官方文档交叉核验（Guangdong-TencentHunyuan-20230901）；元宝作为 C 端入口承载该已备案模型。the Hunyuan model behind Yuanbao is filed (Guangdong-TencentHunyuan-20230901, cross-verified via Tencent Cloud official docs); Yuanbao is the consumer entry carrying that filed model.

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: 主力是腾讯自研混元（Hy）系列——腾讯官方新闻页有模型迭代记录（Hy3 preview 2026-04-24 发布等）；元宝早期以「混元 + DeepSeek 双模型切换」出名，当前对外宣传已收敛到 Hy 系列，模型选择器需登录后可见，本站未登录核实——以登录后界面实际展示为准。
The main line is Tencent's self-developed Hunyuan (Hy) series — Tencent's official news page has model-iteration records (Hy3 preview released 2026-04-24 etc.); Yuanbao was early on known for "Hunyuan + DeepSeek dual-model switching", but current external promotion has converged onto the Hy series; the model selector is only visible after signing in and this site did not verify while signed in — go by what the signed-in UI actually shows.
运行位置Where it runs: 网页（yuanbao.tencent.com）+ iOS/Android App + PC/Mac 客户端多端；生成在腾讯云端处理，本机不跑模型。与腾讯 ima（知识库工作台）同属腾讯、都基于混元，但定位不同——元宝是通用助手，ima 是个人知识库 RAG。
Web (yuanbao.tencent.com) + iOS/Android apps + PC/Mac clients, multi-surface; generation is processed on Tencent's cloud, no model runs locally. It and Tencent ima (a knowledge-base workbench) both belong to Tencent and both build on Hunyuan but are positioned differently — Yuanbao is a general assistant, ima is a personal knowledge-base RAG.
本地文件Local files: 对话历史与收藏存于账户侧；文档精读支持多格式、单文档可处理超长文本（官方叙事），上传内容存云端账户。本机不跑模型——客户端只是入口。
Chat history and favorites are stored on the account side; deep document reading supports multiple formats and can process ultra-long single documents (official narrative); uploaded content is stored in the cloud account. No model runs locally — the client is just an entry.
关机后的任务Tasks after shutdown: 对话即时返回，无「关机续跑」形态的官方声明；深度思考是会话内推理档位，不是异步任务。
Chats return instantly with no official statement of an "offline continuation" form; deep thinking is an in-session reasoning tier, not an async task.
工具与扩展Tools & extensions: 联网搜索（独家微信公众号/视频号信源是它的差异点）、文档精读、AI 作图、智能体创建、AI 编程（多语言运行即时验证）等产品内能力线；联动腾讯文档/微信读书。MCP 支持未核验到官方声明。
In-product capability lines incl. web search (exclusive WeChat Official Account / Channels sources are its differentiator), deep document reading, AI image generation, agent creation and AI coding (multi-language run-and-verify); integration with Tencent Docs / WeChat Read. No official statement verified for MCP support.
上下文与记忆Context & memory: 对话历史登录同步多端；有「无痕临时对话」（聊完不入历史）；跨会话长期记忆机制未逐条核验。
Chat history syncs across devices when signed in; there is an "incognito temporary chat" (not saved to history after chatting); the cross-session long-term memory mechanism is not verified item by item.
权限与限制Permissions & limits: 账号体系（微信/QQ/手机号登录）；训练用途开关路径未逐条核验。混元模型备案号见 cn 块（官方公示）。
Account system (WeChat/QQ/phone-number login); the training-use switch path not verified item by item. The Hunyuan model filing number is in the cn block (official disclosure).
适合什么任务Best for: 适合：泡微信生态的人（搜公众号/视频号、评论区 @ 解读、联动腾讯文档与微信读书）、长文档精读、免费额度。不适合：要本机跑模型、要脱离腾讯账号体系。
Suits: people living in the WeChat ecosystem (searching Official Accounts/Channels, @ to interpret in comment sections, integrating Tencent Docs and WeChat Read), long-document deep reading, free quota. Does not suit: running models locally, or leaving the Tencent account system.

头号误解Top misconception

把「元宝」与「混元」混为一谈——混元是模型（hunyuan.tencent.com 是模型/研究站，不能聊天），元宝是搭在其上的 C 端入口Confusing Yuanbao with Hunyuan — Hunyuan is the model (hunyuan.tencent.com is a model/research site you cannot chat on); Yuanbao is the consumer entry built on it
把「双模型自由切换」当现状——当前官方宣传已收敛到 Hy 系列，DeepSeek 是否仍可选需登录后界面确认（本站未登录核实）Treating 'free dual-model switching' as current state — official promotion has converged onto the Hy series; whether DeepSeek remains selectable needs a logged-in check (not verified logged in)
把元宝与腾讯 ima 当同一产品——同基于混元但一个是通用助手、一个是个人知识库工作台Treating Yuanbao and Tencent ima as one product — both run on Hunyuan, but one is a general assistant and the other a personal knowledge-base workbench

未知项清单Known unknowns

当前可选模型清单（登录后可见，未登录核实）
训练用途开关路径与默认态（未逐条核验）
跨会话长期记忆机制（未逐条核验）
文档精读格式/长度上限（未做实测）

The current list of selectable models (visible after signing in, not verified while signed in)
The training-use switch path and default state (not verified item by item)
The cross-session long-term memory mechanism (not verified item by item)
Deep-reading format/length limits (not measured)

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	腾讯元宝 · 官网（产品与能力叙事）Tencent Yuanbao · official site (product and capability narrative)	https://yuanbao.tencent.com
official	腾讯元宝 · 用户协议（官方页，2026-10-22 直抓核验）Tencent Yuanbao · User Agreement (official page, direct-fetch verified 2026-10-22)	https://yuanbao.tencent.com/agreement
changelog	腾讯元宝 · News（产品更新入口，changelog）Tencent Yuanbao · News (product-update entry, changelog)	https://yuanbao.tencent.com/news
official	腾讯云 · AI 工作台备案信息（混元大模型备案号交叉核验）Tencent Cloud · AI Workbench filing information (Hunyuan model filing number cross-verification)	https://cloud.tencent.com/document/api/213/2257

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）：

登录后模型选择器是否仍含 DeepSeek（官方宣传已收敛到 Hy，本站未登录核实）
训练用途开关的路径与默认态
文档精读的格式与单文档长度上限（官方叙事有「约 1000 万字」口径，未做实测）

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):

Whether the signed-in model selector still includes DeepSeek (official promotion has converged onto Hy; this site did not verify while signed in)
The training-use switch path and default state
Format and single-document length limits of deep document reading (the official narrative has an "about 10 million characters" claim, not measured)