AI 编程订阅计划（Coding Plan）对比站 —— 看得见数据来源的订阅指南。
🔗 在线地址：https://wjf1.github.io/coding-plan-radar/
覆盖 20 个平台 / 69 个付费档位（其中国内 10 家、国际 10 家；含 1 个已停售平台作为历史记录保留）， 每条数据标注来源与采集日期。
| 功能 | 说明 |
|---|---|
| ⚡ 快速对比表 | 价格、额度口径、状态（正常/促销/停售），支持搜索、国内/国际筛选、多键排序 |
| ☑️ 多选对比 | 勾选 2–4 个平台生成并排对比视图，选择状态写进 URL（超出上限自动截断） |
| 📈 额度倍率 | 月度额度折算价值 ÷ 月价（awesome-coding-plan 口径），MiniMax 88.65× / Claude 63.6× / ChatGPT 21.82× … |
| 🕒 三周期额度 | 每平台 5h / 周 / 月三档额度与倍率，识别「月倍率高但 5h 倍率低」的节奏陷阱 |
| 📉 价格历史趋势 | 每日记录起步价快照（保留 90 天），页面内以纯 SVG 迷你图呈现走势，涨价/降价自动告警 |
| 💰 Token 实时价格榜 | 浏览器直连 models.dev 拉取第一方模型 API 价（$/百万 Token），当前渲染 80 款编码模型，输出价升序 + ⚡性价比标记 |
| 🧮 订阅 vs API 计算器 | 输入月输出量与模型档位（7 档），实时估算按量成本，对比候选订阅给出「订阅还是按量」结论 |
| 🖥️ IDE 订阅榜 | 12 个 $10–20 档 IDE/编辑器订阅横评（Trae / Copilot / Zed / Kiro / Windsurf…） |
| 🎁 白嫖 / 免费额度 | 官方免费档、学生包、免费 API 额度、限时试用；标注额度 / 门槛 / 证据强度，并列出已失效的坑（如 GitHub Models 已下线） |
| 📡 市场动态 | 促销与停售追踪（人工确认）+ 自动发现的线索队列 + 信息源健康看板 + 本站变更记录 |
| 🌐 中英双语 | 页面右上角一键切换中/英：首屏与全部结构标签（章节标题、表格表头、工具条、计算器标签、搜索占位符）随语言切换，长文正文与导航仍为中文；语言选择随 URL 一起保留 |
| 🔗 链接可分享 | 筛选、排序、搜索词、对比选中的平台、语言全部同步到 URL，复制链接即可复现同一视图 |
| 📤 CSV 导出 | 对比表一键导出 CSV（含换行、引号的安全转义，可直接进 Excel） |
| ♿ 无障碍 | 语义化表格 + ARIA 标签（aria-pressed 与选中态同步）、键盘可达的「跳到主内容」链接、移动端触摸滚动 |
| 🛡️ 来源分级 | 官方直采（绿）＞ 实时数据源（青）＞ 聚合参考（蓝）＞ 社区情报（黄，仅作线索，不进价格表） |
仓库每天 北京时间 09:00（cron 0 1 * * * UTC）自动运行 daily-update.yml。
四个采集任务互相独立、任一失败不影响其余（continue-on-error），最后汇入一个 publish 作业：
scripts/update-snapshot.mjs—— 重新抓取 models.dev，重新生成js/snapshot.js兜底快照scripts/check-pages.mjs—— 按data/sources.json对官方定价页/文档做内容哈希变动检测- 有变动 → 写入
data/alerts.json，站点顶部自动挂「待核实」横幅并开 issue - 页面回到基线哈希时自动 resolved；抓取失败不再静默跳过，而是计入源健康
- 有变动 → 写入
scripts/fetch-feeds.mjs—— 抓取官方 changelog / 状态页 / 社区订阅源，归一化为data/signals.json线索scripts/diff-prices.mjs—— 对比 models.dev / OpenRouter / LiteLLM：价格变动、免费模型增删、价格源分歧publish作业（依赖以上四项）：写入巡检日期 →record-history.mjs记录价格历史快照 → 生成待人工处理清单（去重后写入data/reported.json）→ 提交推送（触发 Pages 重新发布）→ 有事项时开 issue
诚实原则：
- 机器只负责发现，所有价格与促销数字一律人工确认后才进对比表；机器线索单独陈列
- 每个信息源的抓取成败写入
data/sourcehealth.json，在站点「市场动态 → 信息源健康」公开可见， 连续 3 天失败即标注「已失效」，不再假装巡检成功 - JS 渲染的空壳页面（Qoder / CodeBuddy / 火山方舟计费页）与对机器人返回验证页的站点（OpenAI 定价页实测 403） 明确标注为不可自动监控，不做假巡检
- 变动需二次确认：页面内容变化要「下一次巡检仍是同一个新值」才告警（真实变动因此晚一天）， 避免动态渲染页面天天误报
- 不稳定页面自动停用预警：cursor.com/pricing 实测两次抓取正文长度即不同（渲染变体跳变）， 连续确认后回滚达阈值即标记为「需人工核对」并停止自动告警——宁可承认监控不了，也不刷假警报
- 打开 issue 中列出的官方页面，核对新价格 / 新额度
- 更新本地
js/data.js对应平台条目（含新的采集日期） - 将
data/alerts.json中该条目的"resolved": false改为true（或等着它自动回滚解决） - 提交推送，横幅自动消失
- 看站点「市场动态 → 自动发现的线索」，或 issue 里的线索列表
- 人工核实后，促销/停售写进
data/promos.json，白嫖/免费额度写进data/freebies.json - 社区情报（黄标）默认不录入，除非能追到官方出处
├── index.html # 页面骨架（对比表 / Token 榜 / 计算器 / IDE 榜 / 白嫖 / 动态 / FAQ）
├── css/styles.css # 全部样式（深色主题，无框架）
├── js/
│ ├── data.js # ⭐ 订阅计划数据 + IDE 表 + 站内变更记录（日常改这里）
│ ├── app.js # 渲染与交互逻辑（表格/卡片/计算器/实时榜/对比/CSV/多语言/价格趋势/巡检）
│ └── snapshot.js # models.dev 兜底快照（每日自动重新生成，勿手改）
├── data/
│ ├── sources.json # ⭐ 信息源声明清单（唯一事实来源：URL / 分级 / 关键词 / 启用状态与原因）
│ ├── promos.json # 人工确认的促销 / 停售记录（驱动「市场动态」时间线）
│ ├── freebies.json # 人工维护的白嫖 / 免费额度条目
│ ├── alerts.json # 「待核实」预警状态（人工 resolved，或自动回滚解决）
│ ├── price-history.json # 起步价每日快照（保留 90 天，自动维护，驱动价格趋势图）
│ ├── reported.json # 已进入待处理清单的线索去重表（自动维护，避免 issue 重复列同一条）
│ ├── signals.json # 机器发现的线索（90 天滚动窗口，自动维护）
│ ├── sourcehealth.json # 每源抓取成败（自动维护，站点据此显示「已失效」）
│ ├── pagehash.json # 官方页内容哈希基线（自动维护）
│ ├── pricebase.json # 价格 / 免费模型基线（自动维护，用于差异检测）
│ └── meta.json # 最近巡检日期（自动维护，页脚与「信息源健康」据此显示）
├── scripts/
│ ├── lib.mjs # 公用：带 UA/超时抓取、JSON 读写、源健康、极简 RSS 解析、关键词匹配
│ ├── update-snapshot.mjs # 兜底快照
│ ├── check-pages.mjs # 官方页哈希巡检 + 源健康
│ ├── fetch-feeds.mjs # RSS / JSON 源 → 线索
│ ├── diff-prices.mjs # 价格库与免费模型差异检测
│ ├── record-history.mjs # 记录每日起步价快照（90 天滚动）+ 调价告警
│ └── publish-via-api.mjs # 走 GitHub API 发布（github.com 被阻断时替代 git push）
├── docs/screenshots/ # README 配图
└── .github/workflows/
└── daily-update.yml # 定时任务（cron 09:00 北京时间，可手动触发）
# 本地预览（任选其一）
python -m http.server 8765 # http://127.0.0.1:8765
npx serve .
# 数据/内容更新后发布
git add . && git commit -m "update: ..." && git push # Pages 自动重新发布
# 若所在网络阻断了 github.com 的 HTTPS（git push 报 connection reset），改用 API 发布：
node scripts/publish-via-api.mjs "update: ..." # 需要 GITHUB_TOKEN 或已登录的 gh CLI
# 手动跑一次完整巡检
node scripts/update-snapshot.mjs && node scripts/check-pages.mjs \
&& node scripts/fetch-feeds.mjs && node scripts/diff-prices.mjs \
&& node scripts/record-history.mjs无构建步骤、零 npm 依赖（含 RSS 解析，未引入 YAML / XML 库）、纯静态 —— fork 后开启 Pages 即可获得自己的实例。
注：
data/sources.json里的社区源（LINUX DO / V2EX 等）在中国大陆线路不可达，只在 GitHub Actions 的海外出口能抓到；本地跑时它们会正常报失败并记录到sourcehealth.json，这属预期行为。
| 层级 | 来源 | 用途 |
|---|---|---|
| 官方直采 | claude.com/pricing · docs.github.com Copilot 计划 · cursor.com/pricing · docs.bigmodel.cn · platform.kimi.com · trae.ai · MiniMax · 阿里云百炼免费额度 · 百度智能云 等 | 订阅价格（人工核价）+ 每日哈希巡检 |
| 官方订阅源 | GitHub Changelog · Cursor Changelog · OpenAI News · Claude / OpenAI / GitHub Status | 促销与可用性事件线索（RSS/Atom） |
| 实时数据源 | models.dev（API 直连，CORS 全开放）· OpenRouter 模型表 · LiteLLM 价格表 | Token 价格榜 + 兜底快照 + 免费模型 / 价格差异检测 |
| 方法论 | mahonzhan/awesome-coding-plan（2857★） | 额度倍率 / TPS / 三周期额度 / 坑点 |
| 聚合参考 | codingplan.org | 部分国内平台价格（逐条标注） |
| 社区情报 | Hacker News · 少数派 · IT之家 · awesome-free-llm-apis 提交流 | 线索提示（黄色标注，不直接进价格表） |
关于 LINUX DO / V2EX（2026-09-14 结论）：这两个一手性最强的中文社区源，经双环境实测均无法自动获取 —— 大陆线路超时，GitHub Actions 海外出口分别返回
403 Cloudflare与HTTP 200 空响应体。 这说明拦截发生在数据中心 IP 信誉层而非网络封锁层，自建 VPS / RSSHub 同样解决不了（能解决的是住宅代理）。 因此本站停用这两个源并改为人工巡览，不做"看起来在监控"的假象。
| 类型 | 原因 |
|---|---|
| 逆向 / 公益 / API 中转站（free-one-api、chatanywhere 等） | 违反上游 ToS，存在密钥泄露与封号风险；且项目寿命不可预期（已有多个知名仓库 404 消失） |
| 共享账号 / 合租 | 账号安全与平台条款风险 |
| 需登录 cookie 抓取（小红书 / 即刻 / 公众号中转） | 合规与稳定性均不可接受 |
| X / Twitter、36氪 / 机器之心原生 RSS | 前者路由需多组鉴权 token 且官方实例已关闭，后者实测返回反爬页或非 feed |
| 中文羊毛聚合站（福利吧 / hostloc / 什么值得买首页） | 实测主动断连 / 需邀请码 / 与 AI 订阅无关内容为主，噪声比极差 |
| LINUX DO / V2EX（暂缓） | 双环境实测均不可自动获取（403 Cloudflare / 空响应体），属数据中心 IP 信誉拦截，自建 VPS 无效；保留配置但 enabled: false，有住宅代理时可一键启用 |
LiteLLM 只做参考基线，不告警：按模型名匹配会把不同 SKU 配到一起（如 litellm 的
azure/gpt-5.6转售价 $30 vs 第一方openai/gpt-5.6$20），首个真实 CI 运行即产生 146 条假"价格分歧"信号。 因此它只抓取、只记录基线供人工比对，不再自动报警。
2026-09-14：原「B站每日播报」来源已移除。相关条目的原始出处仅剩视频、无法结构化核验， 故改由上述官方 / 结构化源持续找官方确认；未能复核的条目在站点上明确标注「⚠ 待重新核实」。
- 价格与额度可能随时变化，付款前务必以平台官网为准
- 「额度倍率」「编辑推荐」为基于公开数据的社区口径估算与主观参考，非官方承诺，非广告
- 免费额度条目的领取条件与规则以官方为准，本站仅做汇总与出处标注
- 本站不含任何推广链接，与所列平台无商业关系；所有商标归各自所有者
A comparison site for AI coding subscription plans (Coding Plan) — a subscription guide where you can see where every number comes from.
🔗 Live: https://wjf1.github.io/coding-plan-radar/
Covers 20 platforms / 69 paid tiers (10 China-region, 10 international, plus 1 discontinued platform kept as a historical record). Every data point is annotated with its source and collection date.
| Feature | Description |
|---|---|
| ⚡ Quick comparison table | Price, quota definition, status (active / promo / discontinued); search, China vs. international filter, multi-key sorting |
| ☑️ Multi-select comparison | Tick 2–4 platforms to build a side-by-side comparison view; the selection is stored in the URL (extra entries are truncated) |
| 📈 Quota ratio | Monthly quota value ÷ monthly price, using the awesome-coding-plan methodology — MiniMax 88.65× / Claude 63.6× / ChatGPT 21.82× … |
| 🕒 Three-period quotas | Per-platform 5h / weekly / monthly quota and ratios, to expose the "high monthly ratio but low 5h ratio" pacing trap |
| 📉 Price history trend | Daily starting-price snapshots (90-day retention) rendered as inline SVG sparklines, with automatic up/down price-change alerts |
| 💰 Live Token price board | Fetched directly in the browser from models.dev ($/million tokens); currently renders 80 coding models, sorted by output price with a ⚡ value flag |
| 🧮 Subscription vs. API calculator | Enter monthly output volume and an API model tier (7 tiers) to estimate pay-as-you-go cost and decide "subscribe or pay per token" |
| 🖥️ IDE subscription board | 12 IDE/editor subscriptions in the $10–20 band (Trae / Copilot / Zed / Kiro / Windsurf…) |
| 🎁 Free tiers & freebies | Official free tiers, student packs, free API quotas, limited-time trials — each with quota / eligibility / evidence strength, plus retired offers (e.g. GitHub Models is gone) |
| 📡 Market dynamics | Promo and discontinuation tracking (human-confirmed) + an auto-discovered lead queue + a source-health dashboard + site changelog |
| 🌐 Bilingual UI | One-click Chinese / English switch in the header: the hero and all structural labels (section titles, table headers, toolbars, calculator labels, search placeholders) follow the language; long-form prose and the nav stay Chinese. The choice persists in the URL |
| 🔗 Shareable links | Filters, sorting, search terms, compared platforms and language all sync to the URL — copy the link to reproduce the exact view |
| 📤 CSV export | One-click CSV export of the comparison table (safe escaping of newlines and quotes, Excel-ready) |
| ♿ Accessibility | Semantic tables with ARIA labels (aria-pressed kept in sync with selection state), a keyboard-reachable "skip to content" link, touch scrolling on mobile |
| 🛡️ Source tiers | Official direct (green) > real-time data source (cyan) > aggregated reference (blue) > community intel (yellow, leads only — never enters the price table) |
Every day at 09:00 Beijing time (cron 0 1 * * * UTC) the repo runs daily-update.yml.
Four collection jobs run independently — one failing does not affect the others (continue-on-error) — then a publish job ties them together:
scripts/update-snapshot.mjs— refetch models.dev and regenerate thejs/snapshot.jsfallback snapshotscripts/check-pages.mjs— content-hash change detection on official pricing/documentation pages, driven bydata/sources.json- On change → write to
data/alerts.json, show a "pending verification" banner on the site and open an issue - Auto-resolves when a page returns to its baseline hash; fetch failures are no longer silently skipped — they count against source health
- On change → write to
scripts/fetch-feeds.mjs— fetch official changelogs / status pages / community feeds and normalize them into leads indata/signals.jsonscripts/diff-prices.mjs— compare models.dev / OpenRouter / LiteLLM: price moves, free-model additions and removals, disagreements between price sourcespublishjob (depends on the four above): write the inspection date → record price-history snapshots viarecord-history.mjs→ build the human review list (deduplicated intodata/reported.json) → commit and push (triggers a Pages redeploy) → open an issue when there is something to review
Honesty principles
- Machines only discover; every price and promo number is human-confirmed before it enters the comparison table. Machine leads are listed separately.
- Per-source fetch success/failure is written to
data/sourcehealth.jsonand shown publicly under "Market dynamics → Source health". Three consecutive days of failure marks a source as dead — no pretending the inspection succeeded. - JavaScript-only shell pages (Qoder / CodeBuddy / 火山方舟 billing pages) and sites that serve a challenge page to bots (OpenAI's pricing page returns 403) are explicitly labelled as not monitorable — no fake monitoring.
- Changes require double confirmation: a changed page must still show the same new value on the next run before it alerts (so a real change is reported one day late). This prevents dynamic pages from generating false alarms every day.
- Unstable pages are automatically dropped from alerting: cursor.com/pricing returns a different body length on two consecutive fetches (rendering variants jump around). Once confirmed, it is marked "needs manual review" and alerting stops — better to admit a page cannot be monitored than to spam false alarms.
- Open the official page listed in the issue and verify the new price / quota
- Update the matching platform entry in
js/data.js(including a new collection date) - Set
"resolved": falsetotrueindata/alerts.jsonfor that entry (or wait for the automatic rollback to resolve it) - Commit and push — the banner disappears
- Look at "Market dynamics → Auto-discovered leads" on the site, or the lead list in the issue
- After verifying by hand, write promos/discontinuations into
data/promos.jsonand freebies intodata/freebies.json - Community intel (yellow) is not entered by default unless an official source can be traced
├── index.html # Page skeleton (comparison table / token board / calculator / IDE board / freebies / dynamics / FAQ)
├── css/styles.css # All styling (dark theme, no framework)
├── js/
│ ├── data.js # ⭐ Subscription plan data + IDE table + in-site changelog (edit this day to day)
│ ├── app.js # Rendering and interaction (tables/cards/calculator/live board/compare/CSV/i18n/price trend/inspection)
│ └── snapshot.js # models.dev fallback snapshot (regenerated daily — do not edit by hand)
├── data/
│ ├── sources.json # ⭐ Source manifest (single source of truth: URLs / tiers / keywords / enabled state and reasons)
│ ├── promos.json # Human-confirmed promos and discontinuations (drives the "market dynamics" timeline)
│ ├── freebies.json # Human-maintained free-tier / freebie entries
│ ├── alerts.json # "Pending verification" alert state (resolved by hand, or by automatic rollback)
│ ├── price-history.json # Daily starting-price snapshots (90-day retention, auto-maintained, drives the price trend chart)
│ ├── reported.json # Dedup table of leads already surfaced for review (auto-maintained, keeps issues from repeating)
│ ├── signals.json # Machine-discovered leads (90-day rolling window, auto-maintained)
│ ├── sourcehealth.json # Per-source fetch success/failure (auto-maintained; drives the "dead" badge)
│ ├── pagehash.json # Official page content-hash baselines (auto-maintained)
│ ├── pricebase.json # Price / free-model baselines (auto-maintained, used for diffing)
│ └── meta.json # Last inspection date (auto-maintained; shown in the footer and "source health")
├── scripts/
│ ├── lib.mjs # Shared helpers: fetch with UA/timeout, JSON IO, source health, minimal RSS parser, keyword matching
│ ├── update-snapshot.mjs # Fallback snapshot
│ ├── check-pages.mjs # Official page hash inspection + source health
│ ├── fetch-feeds.mjs # RSS / JSON sources -> leads
│ ├── diff-prices.mjs # Price-base and free-model diffing
│ ├── record-history.mjs # Record daily starting-price snapshots (90-day rolling) + price-change alerts
│ └── publish-via-api.mjs # Publish through the GitHub API (fallback when github.com is blocked)
├── docs/screenshots/ # README images
└── .github/workflows/
└── daily-update.yml # Scheduled job (cron 09:00 Beijing time, manually triggerable)
# Local preview (pick one)
python -m http.server 8765 # http://127.0.0.1:8765
npx serve .
# Publishing after data/content changes
git add . && git commit -m "update: ..." && git push # Pages redeploys automatically
# If your network blocks github.com over HTTPS (git push reports connection reset), publish via the API:
node scripts/publish-via-api.mjs "update: ..." # needs GITHUB_TOKEN or a logged-in gh CLI
# Run one full inspection by hand
node scripts/update-snapshot.mjs && node scripts/check-pages.mjs \
&& node scripts/fetch-feeds.mjs && node scripts/diff-prices.mjs \
&& node scripts/record-history.mjsNo build step, zero npm dependencies (including the RSS parser — no YAML/XML library), purely static — fork it, enable Pages, and you have your own instance.
Note: the community sources in
data/sources.json(LINUX DO / V2EX and friends) are unreachable from mainland China routes and can only be fetched from GitHub Actions' overseas egress. Running locally, they will report failure and be recorded insourcehealth.json— that is expected behaviour.
| Tier | Source | Used for |
|---|---|---|
| Official direct | claude.com/pricing · docs.github.com Copilot plans · cursor.com/pricing · docs.bigmodel.cn · platform.kimi.com · trae.ai · MiniMax · Alibaba Cloud Model Studio free quota · Baidu AI Cloud etc. | Subscription prices (human-verified) + daily hash inspection |
| Official feeds | GitHub Changelog · Cursor Changelog · OpenAI News · Claude / OpenAI / GitHub status | Promo and availability event leads (RSS/Atom) |
| Real-time sources | models.dev (direct API, CORS fully open) · OpenRouter models · LiteLLM price table | Token price board + fallback snapshot + free-model / price diffing |
| Methodology | mahonzhan/awesome-coding-plan (2,857★) | Quota ratios / TPS / three-period quotas / pitfalls |
| Aggregated reference | codingplan.org | Some China-region platform prices (annotated per entry) |
| Community intel | Hacker News · sspai · IT之家 · awesome-free-llm-apis submissions | Lead hints (yellow labels, never enter the price table directly) |
On LINUX DO / V2EX (conclusion, 2026-09-14): the two strongest first-hand Chinese community sources cannot be fetched automatically in either environment — mainland routes time out, while GitHub Actions' overseas egress gets
403 CloudflareandHTTP 200 with an empty bodyrespectively. The block therefore happens at the datacenter IP reputation layer, not the network-blocking layer, so a self-hosted VPS or RSSHub cannot fix it (a residential proxy can). The site therefore disables both sources and falls back to manual browsing, rather than putting on a show of monitoring them.
| Type | Reason |
|---|---|
| Reverse-engineered / public-good / API relay services (free-one-api, chatanywhere, etc.) | Violate upstream ToS, with key-leak and ban risks; project lifespan is unpredictable (several well-known repos have 404'd) |
| Shared or co-rented accounts | Account security and platform ToS risk |
| Sources requiring login cookies (Xiaohongshu / Jike / WeChat relay) | Neither compliance nor reliability is acceptable |
| X / Twitter, 36Kr / Synced native RSS | The former needs multiple auth tokens and its official instance is shut; the latter returns anti-bot pages or non-feeds |
| Chinese deal aggregators (福利吧 / hostloc / SMZDM homepage) | Measured active disconnects / invite-only / mostly content unrelated to AI subscriptions — a terrible signal-to-noise ratio |
| LINUX DO / V2EX (deferred) | Neither environment can fetch them (403 Cloudflare / empty body); it is datacenter-IP reputation blocking, so a VPS does not help. Configuration is kept with enabled: false so a residential proxy can switch it on in one line |
LiteLLM is a reference baseline only, never alerted on: matching by model name pairs up different SKUs (e.g. litellm's
azure/gpt-5.6resale at $30 vs. first-partyopenai/gpt-5.6at $20). The very first real CI run produced 146 false "price disagreement" signals. It is therefore fetched and recorded as a baseline for human comparison, but no longer alerts automatically.
2026-09-14: the original "daily Bilibili roundup" source was removed. Those entries only existed as video with no structured way to verify them, so confirmation now relies on the official/structured sources above. Entries that could not be re-verified are explicitly marked "⚠ pending re-verification" on the site.
- Prices and quotas can change at any time — always confirm on the vendor's own site before paying
- "Quota ratio" and "editor's picks" are community-methodology estimates and subjective references based on public data — not official promises, not advertising
- Eligibility and rules for free tiers are governed by the vendors; this site only aggregates them and cites sources
- This site contains no affiliate links and has no commercial relationship with the platforms listed; all trademarks belong to their respective owners
📝 变更记录 / Changelog：CHANGELOG.md