← Chat

PoePoe

Quora

多模型聚合Multi-model aggregation 点数制Points-based bot创建Bot creation 海外Outside China 未备案Not filed partial lifecycle: active 核验Checked 2026-10-30

Poe 是「一个账号用多家模型」的聚合对话产品（Quora 出品）：GPT/Claude/DeepSeek/Veo 等经 Poe 侧接入、用户直接选模型，点数制计费，自建 bot 面在册；数据全部经 Poe 服务器中转。

Poe is the "one account, many models" aggregator chat product (by Quora): GPT/Claude/DeepSeek/Veo etc. connect through Poe's side, users pick models directly, points-based billing, and a build-your-own-bot surface is on record; all data is relayed through Poe's servers.

适合与不适合When to use it, when not to

适合：横向对比多家模型、自建 bot 人设、含国产模型接入。不适合：要免费无限量旗舰模型、要大陆直连、要数据不经第三方中转。

Fits: comparing many models side by side, building your own bots, domestic-model access included. Not a fit: unlimited free flagship models, direct mainland access, data not relayed through a third party.

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 2026-10-30

出品属地Origin: 国外出品foreign-made　国外主体（Quora/Poe，美国；网信办公示名单未见对应备案主体）。foreign entity (Quora/Poe, USA; no matching filing entity in the CAC registry).
国内使用Direct access: 直连不可达direct blocked　poe.com 大陆直连 000（网络不可达，2026-10-30）。poe.com measured 000 direct from mainland (network unreachable, 2026-10-30).
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（2026-10-30 检索百炼公示页与批次 PDF 镜像，未见 Poe/Quora 条目）；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 (Bailian disclosure page and batch PDF mirrors searched 2026-10-30; no Poe/Quora entry); 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: 多模型聚合是它的本体形态：官方页当日可见「Chat with the best AI… Explore GPT-5, Claude-Sonnet-5, DeepSeek-R1, Veo-3.1, and thousands of others」——用户直接选模型对话，模型经 Poe 侧接入各家 API（各模型计费点数不同）。
Multi-model aggregation is its native form: the official page on the day shows "Chat with the best AI… Explore GPT-5, Claude-Sonnet-5, DeepSeek-R1, Veo-3.1, and thousands of others" — users pick models and chat directly, models connect via Poe's side to each vendor's API (point costs differ per model).
运行位置Where it runs: 网页（poe.com）+ iOS/Android App；对话经 Poe 服务器转发到各模型 API，本机不跑模型。当日 poe.com 对无头抓取 307→login 403（登录墙+CF，非死链）。
Web (poe.com) + iOS/Android apps; chats are relayed through Poe's servers to each model API, nothing runs locally. poe.com returned 307→login 403 to headless fetching on the day (login wall + CF, not a dead link).
本地文件Local files: 对话历史存于账户侧、登录同步；未见官方声明的本机数据面——所有模型调用都经 Poe 服务器中转，对话内容必然过 Poe 侧（这是聚合中转产品的结构性前提，不是可关掉的选项）。
Chat history stored on the account side, synced after login; no locally-held data surface declared by the vendor — every model call is relayed through Poe's servers, so conversation content necessarily passes the Poe side (a structural premise of an aggregation-relay product, not an optional toggle).
关机后的任务Tasks after shutdown: 对话即时返回；异步任务形态未核验到官方声明。点数制对高频/长上下文模型有用量限制。
Chat returns instantly; no official statement verified for an async-task form. The points system limits usage on high-frequency/long-context models.
工具与扩展Tools & extensions: 官方页当日可见群聊（group chat）、Leaderboard、bot 创建面（自建 bot 挂模型+人设）；图像/视频模型（Veo-3.1 在册）也在聚合清单内。MCP 支持未核验到官方声明。
The official page on the day shows group chat, Leaderboard, and a bot-creation surface (build your own bot on a model + persona); image/video models (Veo-3.1 on record) are also in the aggregation list. No MCP support verified in official statements.
上下文与记忆Context & memory: 会话按对话组织、登录同步，bot 人设（自建 bot 挂模型+人设）是跨会话复用的持久配置面；模型侧的上下文窗口长度按各家模型各异，Poe 产品层的跨会话记忆机制未核验到官方声明。
Chats organized per conversation, synced after login; bot personas (self-built bots binding a model + persona) are a persistent configuration surface reused across sessions. Context-window length varies per underlying model; Poe's product-level cross-session memory mechanism has no verified official statement.
权限与限制Permissions & limits: 账号体系（注册登录）；点数制是用量权限面；训练用途开关路径未逐条核验—— Poe 曾以「用户对话用于训练」条款引发争议，要按官方 Terms/Privacy 现行文本逐项看。
Account system (register + login); the points system is the usage-permission surface; training-usage toggle path not verified item by item — Poe previously drew controversy over "user chats used for training" terms; read the current official Terms/Privacy item by item.
适合什么任务Best for: 适合：一个账号横向用多家模型（含国产 DeepSeek 系）、要自建 bot 人设。不适合：要免费无限量用旗舰模型（点数制）、要大陆直连（未备案）、要数据不经 Poe 中转。
Fits: using many vendors' models under one account (domestic DeepSeek models included), building your own bot personas. Not a fit: unlimited free flagship models (points system), direct mainland access (not filed), data not relayed through Poe.

头号误解Top misconception

Poe 是中转聚合层——你的对话经 Poe 服务器转发到各家模型 API，敏感数据要按「多一方经手」评估Poe is a relay/aggregation layer — your chats are forwarded from Poe's servers to each model API; assess sensitive data as "one more party handling it"
点数制不是「免费无限」——旗舰模型与长上下文消耗快，免费档边界要实测The points system is not "free unlimited" — flagship models and long contexts burn points fast; the free-tier boundary needs measuring
模型名带后缀（如 Claude-Sonnet-5）是 Poe 侧接入的具体版本——别把它等同于该厂最新能力面Model names with suffixes (e.g. Claude-Sonnet-5) are the specific versions Poe connects on its side — do not equate them with that vendor's latest capability

未知项清单Known unknowns

点数汇率与免费档边界（未逐条核验）
训练用途开关路径与默认态（未逐条核验）
跨会话记忆机制（未核验到官方声明）
聚合层对各模型上下文的处理（未核验）

Point exchange rate and free-tier boundary (not verified item by item)
Training-usage toggle path and default state (not verified item by item)
Cross-session memory mechanism (no official statement verified)
How the aggregation layer handles each model's context (unverified)

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 2026-10-30。

类型Type	名称Name	链接Link
official	Poe · 官网（产品定位与模型聚合叙事，2026-10-30 代理抓 307→login 403=登录墙非死链）Poe · official site (product positioning and model-aggregation narrative, proxy-fetched 307→login 403 on 2026-10-30 = login wall, not a dead link)	https://poe.com
official	Poe · Terms（官方条款，2026-10-30 代理抓 200）Poe · Terms (official terms, proxy-fetched 200 on 2026-10-30)	https://poe.com/terms
official	Poe · Privacy（官方隐私页，2026-10-30 代理抓 200）Poe · Privacy (official privacy page, proxy-fetched 200 on 2026-10-30)	https://poe.com/privacy-policy
changelog	Poe · Changelog（官方更新线，changelog，2026-10-30 代理抓 200）Poe · Changelog (official update line, changelog, proxy-fetched 200 on 2026-10-30)	https://poe.com/changelog

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）：

各模型的点数消耗与免费档边界（点数制的具体汇率官方页不直白）
训练用途开关路径与默认态（Terms/Privacy 现行文本逐项看）
聚合层的上下文截断策略（经中转后各模型上下文是否打折）

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):

Point costs per model and the free-tier boundary (the points exchange rate is not spelled out on official pages)
Training-usage toggle path and default state (read the current Terms/Privacy item by item)
Context truncation policy at the aggregation layer (whether each model's context is discounted after relaying)