选型 · Pro / Pro+ / Ultra、同类与边界Choose · Pro, Pro+, Ultra, Rivals, and Boundaries.
选档不是看价格，是看你撞的是哪堵墙：额度、模型、Cloud Agents、Bugbot，还是团队管控。这一章把 Pro、Pro+、Ultra、Teams、Enterprise 和计费方式（Auto + Composer 池 vs 前沿模型 API 池）摆开，再划三条线：Cursor 和 Claude Code / Codex / Copilot / Windsurf / Warp 的分工，个人 builder 的组合，以及最该记住的那条 —— Cursor 的 Computer Use boundary 不是通用电脑控制。Choosing a tier isn't about price — it's about which wall you hit: quota, models, Cloud Agents, Bugbot, or team controls. This chapter lays out Pro, Pro+, Ultra, Teams, Enterprise and billing (Auto + Composer pool vs frontier-model API pool), then draws three lines: Cursor vs Claude Code / Codex / Copilot / Windsurf / Warp, the personal-builder tool stack, and the Computer Use boundary.
选档不是看价格，是看你撞的是哪堵墙：额度、模型、Cloud Agents、Bugbot，还是团队管控。Choosing a tier isn't about price — it's about which wall you hit: usage, models, Cloud Agents, Bugbot, or team controls.这一章把 Pro、Pro+、Ultra、Teams、Enterprise 和计费方式摆开，再划三条线：Cursor 与 Claude Code / Codex / Copilot / Windsurf / Warp 的分工、个人 builder 的工具组合、以及最该记住的那条 —— Cursor 的 Computer Use boundary 不是通用电脑控制。这是全书的地图章：它在哪儿停、该交给谁接。This chapter lays out Pro, Pro+, Ultra, Teams, Enterprise, and billing, then draws three lines: Cursor vs Claude Code / Codex / Copilot / Windsurf / Warp, the personal-builder tool stack, and the Computer Use boundary. This is the book's map chapter: where it stops, and who picks up from there.
— I
先认清你撞的是哪堵墙Name the Wall You Keep Hitting First.
每一档解开的不是笼统的「更高级」，是一堵具体的墙。What each tier unlocks isn't a vague 'more premium' — it's a specific wall.
档位Tier
价格（截至 2026-07-10）Price (as of 2026-07-10)
它解开的那堵墙The wall it unlocks
Hobby
免费Free
够试用，Agent 与 Tab 很快见底Enough to try; Agent and Tab run out fast
Pro
$20/月$20/mo
全部功能 + $20 API 用量All features + $20 of API usage
Pro+
$60/月
更高 API 用量，适合日常 Agent 用户More API usage, suited to daily Agent users
Ultra
$200/月
最高个人用量，适合多 agent / 自动化重度用户Highest individual usage, suited to multi-agent / automation power users
计费的关键不在档位，在两套额度池：Auto、Composer 2.5、Grok 4.5 走「第一方模型池」（宽得多，但不是无限），手动选 Claude / GPT / Gemini 这些前沿模型走 API 那套（烧得快）；额度用完转 on-demand，按模型 API 价走、质量速度不降；Max 模式按 token 算。1注 1Note 1Cursor · 套餐与价格 —— Hobby 免费、Pro $20/月、Pro+ $60/月、Ultra $200/月、Teams Standard $40/座·月、Teams Premium $120/座·月（Standard 的 5× Agent 用量）、Enterprise 定制。各档含一份模型用量额度（Pro 含 $20 API 用量、Pro+ $70、Ultra $400）。截至 2026-07-10，Cursor 改价很勤，引用前以官网为准。Cursor · Pricing — Hobby free, Pro $20/mo, Pro+ $60/mo, Ultra $200/mo, Teams Standard $40/seat·mo, Teams Premium $120/seat·mo (5x Standard's Agent usage), Enterprise custom. Each plan includes a model-usage allowance ($20 of API usage on Pro, $70 on Pro+, $400 on Ultra). As of 2026-07-10; Cursor reprices often, so trust the site before citing.2注 2Note 2Cursor Docs · Models & Pricing —— 个人档两套额度池：「第一方模型池」（Auto、Composer 2.5、Grok 4.5 —— Grok 4.5 由 Cursor 与 SpaceXAI 联合训练；included 用量宽得多，但非无限）与 API 池（手动选的前沿模型按该模型 API 价）；超出转 on-demand，官方明言「质量与速度不降级」；Max 按 token 的模型 API 价计费（旧 request 制套餐加 20%）；长上下文加价按模型而异；付费个人档 Tab 补全不限量。截至 2026-07-10。Cursor Docs · Models & Pricing — two pools on individual plans: the First-party models pool (Auto, Composer 2.5, Grok 4.5 — the latter jointly trained by Cursor and SpaceXAI; significantly more included usage, but not unlimited) and the API pool (hand-picked frontier models at each model's API rate); past the allowance you go on-demand, with requests 'never downgraded in quality or speed'; Max bills per token at the model's API rate (legacy request-based plans add 20%); long-context surcharges vary by model; Tab completions are unlimited on paid individual plans. As of 2026-07-10.The key to billing isn't the tier — it's the two pools: Auto, Composer 2.5, and Grok 4.5 ride the First-party pool (much wider, but not unlimited), while manually picking a frontier model (Claude / GPT / Gemini) rides the API pool (burns fast); past your allowance you go on-demand at the model's API rate with no drop in quality or speed; Max mode bills per token.1注 1Note 1Cursor · 套餐与价格 —— Hobby 免费、Pro $20/月、Pro+ $60/月、Ultra $200/月、Teams Standard $40/座·月、Teams Premium $120/座·月（Standard 的 5× Agent 用量）、Enterprise 定制。各档含一份模型用量额度（Pro 含 $20 API 用量、Pro+ $70、Ultra $400）。截至 2026-07-10，Cursor 改价很勤，引用前以官网为准。Cursor · Pricing — Hobby free, Pro $20/mo, Pro+ $60/mo, Ultra $200/mo, Teams Standard $40/seat·mo, Teams Premium $120/seat·mo (5x Standard's Agent usage), Enterprise custom. Each plan includes a model-usage allowance ($20 of API usage on Pro, $70 on Pro+, $400 on Ultra). As of 2026-07-10; Cursor reprices often, so trust the site before citing.2注 2Note 2Cursor Docs · Models & Pricing —— 个人档两套额度池：「第一方模型池」（Auto、Composer 2.5、Grok 4.5 —— Grok 4.5 由 Cursor 与 SpaceXAI 联合训练；included 用量宽得多，但非无限）与 API 池（手动选的前沿模型按该模型 API 价）；超出转 on-demand，官方明言「质量与速度不降级」；Max 按 token 的模型 API 价计费（旧 request 制套餐加 20%）；长上下文加价按模型而异；付费个人档 Tab 补全不限量。截至 2026-07-10。Cursor Docs · Models & Pricing — two pools on individual plans: the First-party models pool (Auto, Composer 2.5, Grok 4.5 — the latter jointly trained by Cursor and SpaceXAI; significantly more included usage, but not unlimited) and the API pool (hand-picked frontier models at each model's API rate); past the allowance you go on-demand, with requests 'never downgraded in quality or speed'; Max bills per token at the model's API rate (legacy request-based plans add 20%); long-context surcharges vary by model; Tab completions are unlimited on paid individual plans. As of 2026-07-10.所以判断很简单：大多数活让 Auto 跑，Pro 的池子就够；天天手动用前沿模型、或开 Max 处理大上下文，才升 Pro+ / Ultra；需要 SSO、admin、团队共享上下文，才上 Teams。Pro 到 Ultra 解开的是额度，不是能力 —— 功能 Pro 就全给了。别为「更高级」付钱，为你真撞的那堵墙付钱。So the call is simple: let Auto run most work and Pro's pool covers you; only daily frontier-model use, or Max for big context, justifies Pro+ / Ultra; only SSO, admin, and shared team context justify Teams. Pro to Ultra unlocks quota, not capability — Pro already gives you every feature. Don't pay for 'more premium' — pay for the wall you actually hit.
自测问题Self-test question
如果是If yes
如果不是If no
你每周是否多次撞到 included usage / on-demand？Do you hit included usage / on-demand multiple times per week?
看 Pro+ / UltraConsider Pro+ / Ultra
留在 ProStay on Pro
你是否每天跑 Agent，多次开 Max 或指定前沿模型？Do you run Agent daily and often use Max or named frontier models?
Pro+ 起步，重度看 UltraStart at Pro+; power use may justify Ultra
Auto + Composer 优先Prefer Auto + Composer
你是否需要多人账单、SSO、隐私模式统一执行？Need team billing, SSO, enforced Privacy Mode?
Teams
个人档即可Individual plan is enough
你是否需要 SCIM、池化用量、模型/MCP/网络限制、审计？Need SCIM, pooled usage, model/MCP/network restrictions, audit?
Enterprise
不要用 Enterprise 解决个人额度焦虑Do not use Enterprise to solve personal quota anxiety
你还没有稳定使用 Agent，只是 Tab / Cmd+K 多：不需要升级。You do not use Agent steadily yet, mostly Tab / Cmd+K: do not upgrade.
你多数任务能用 Auto / Composer 跑完：不需要升级。Most tasks finish on Auto / Composer: do not upgrade.
你只是想偶尔试强模型：先用 on-demand 看账单。You only want occasional frontier models: try on-demand before upgrading.
你没有 review 产能审多个并行 agent：不要因为 Cloud Agents 升级。You cannot review many parallel agents: do not upgrade for Cloud Agents alone.
你每天让 Agent 做多文件任务，且经常被额度打断。You run multi-file Agent tasks daily and often hit limits.
你频繁用 Max / 前沿模型处理大上下文、复杂重构、长链路 debug。You often use Max / frontier models for large context, refactors, long debugging chains.
你并行派 Cloud Agents，有稳定测试和 PR review 流程。You dispatch Cloud Agents in parallel and have stable tests plus PR review.
你把 CLI / automations 放进真实工作流，而不是偶尔玩一次。CLI / automations are in your real workflow, not an occasional experiment.
所以「什么烧钱」值得记牢：同一个任务，agent 在内部要多次调模型（不是一问一答），一个 agent 任务因此比一句 chat 贵得多；长上下文也贵 —— 加价规则按模型而异（有的超过 200k 输入按 2× 计，有的到 1M 明确不加价），开 Max 前看一眼模型表；Max 按模型 API 价走、烧得最快，而 Cloud Agents 和 Automations 一律跑在 Max 上、没有关闭开关。5注 5Note 5Cursor Docs · Cloud Agents —— Cloud Agents 使用一组精选模型、一律跑在 Max Mode，没有关闭开关；Automations 因为跑的就是 cloud agents，同样恒为 Max Mode 计费。截至 2026-07-10。Cursor Docs · Cloud Agents — Cloud Agents use a curated selection of models that always run in Max Mode, with no toggle to turn it off; Automations, since they run as cloud agents, likewise always bill at Max Mode. As of 2026-07-10.反过来，Tab 补全在各付费档不限量、根本不吃这套额度池。2注 2Note 2Cursor Docs · Models & Pricing —— 个人档两套额度池：「第一方模型池」（Auto、Composer 2.5、Grok 4.5 —— Grok 4.5 由 Cursor 与 SpaceXAI 联合训练；included 用量宽得多，但非无限）与 API 池（手动选的前沿模型按该模型 API 价）；超出转 on-demand，官方明言「质量与速度不降级」；Max 按 token 的模型 API 价计费（旧 request 制套餐加 20%）；长上下文加价按模型而异；付费个人档 Tab 补全不限量。截至 2026-07-10。Cursor Docs · Models & Pricing — two pools on individual plans: the First-party models pool (Auto, Composer 2.5, Grok 4.5 — the latter jointly trained by Cursor and SpaceXAI; significantly more included usage, but not unlimited) and the API pool (hand-picked frontier models at each model's API rate); past the allowance you go on-demand, with requests 'never downgraded in quality or speed'; Max bills per token at the model's API rate (legacy request-based plans add 20%); long-context surcharges vary by model; Tab completions are unlimited on paid individual plans. As of 2026-07-10.So 'what burns money' is worth memorizing: for one task the agent calls the model many times internally (not one question, one answer), which makes an agent task far pricier than a single chat; long context costs too — the surcharge varies by model (some bill 2× past 200k input, some explicitly add nothing up to 1M), so check the model table before opening Max; Max runs at the model's API rate and burns fastest, and Cloud Agents and Automations always run on Max with no off switch.5注 5Note 5Cursor Docs · Cloud Agents —— Cloud Agents 使用一组精选模型、一律跑在 Max Mode，没有关闭开关；Automations 因为跑的就是 cloud agents，同样恒为 Max Mode 计费。截至 2026-07-10。Cursor Docs · Cloud Agents — Cloud Agents use a curated selection of models that always run in Max Mode, with no toggle to turn it off; Automations, since they run as cloud agents, likewise always bill at Max Mode. As of 2026-07-10. In the other direction, Tab completions are unlimited on every paid plan and don't draw on this pool at all.2注 2Note 2Cursor Docs · Models & Pricing —— 个人档两套额度池：「第一方模型池」（Auto、Composer 2.5、Grok 4.5 —— Grok 4.5 由 Cursor 与 SpaceXAI 联合训练；included 用量宽得多，但非无限）与 API 池（手动选的前沿模型按该模型 API 价）；超出转 on-demand，官方明言「质量与速度不降级」；Max 按 token 的模型 API 价计费（旧 request 制套餐加 20%）；长上下文加价按模型而异；付费个人档 Tab 补全不限量。截至 2026-07-10。Cursor Docs · Models & Pricing — two pools on individual plans: the First-party models pool (Auto, Composer 2.5, Grok 4.5 — the latter jointly trained by Cursor and SpaceXAI; significantly more included usage, but not unlimited) and the API pool (hand-picked frontier models at each model's API rate); past the allowance you go on-demand, with requests 'never downgraded in quality or speed'; Max bills per token at the model's API rate (legacy request-based plans add 20%); long-context surcharges vary by model; Tab completions are unlimited on paid individual plans. As of 2026-07-10.
— II
同类与配对：取舍，不踩Rivals and Pairing: Trade-Offs, No Bashing.
它们不是谁好谁坏，是各自压在不同的轴上。It's not which is better — it's which axis each one bets on.给取舍轴，不拉踩：Cursor 压在「编辑器集成 + 索引 + Agent + MCP + Cloud Agents」这一轴上；Claude Code 压在 terminal-native / Claude-native 的工程闭环上；Codex 压在 OpenAI / ChatGPT / cloud tasks / PR review 生态上；Copilot 压在 GitHub / VS Code / Microsoft 企业生态上；Windsurf 和 Cursor 同在 IDE + agent 赛道；Warp 则是 AI terminal / shell 工作台。Give the trade-off axis, no bashing: Cursor bets on editor integration + index + Agent + MCP + Cloud Agents; Claude Code on terminal-native / Claude-native engineering loops; Codex on OpenAI / ChatGPT / cloud tasks / PR review; Copilot on GitHub / VS Code / Microsoft enterprise; Windsurf in the IDE + agent lane; Warp as an AI terminal / shell workbench.
工具Tool
强项Strength
什么时候选它Pick it when
Cursor
IDE-native：代码、索引、终端、MCP、Agent、Cloud Agents 在一个工作台，外加 Marketplace 与 SDKIDE-native: code, index, terminal, MCP, Agent, Cloud Agents in one workbench, plus a Marketplace and SDK
项目上下文和可审 diff 最重要Project context and reviewable diffs matter most
同类 IDE + agent，产品取向更面向新手路径Similar IDE + agent, product path often friendlier to beginners
你更喜欢它的交互节奏或团队已有席位You prefer its interaction rhythm or your team already has seats
Warp
AI terminal / shell 工作台AI terminal / shell workbench
工作主要是命令流、运维、跨项目 shell，而不是代码上下文Work is mostly command flow, ops, cross-project shell, not code context
判断：要一个把项目上下文、编辑、终端、review 放在同一窗口的主力环境，Cursor；活主要在 shell 和命令流，Claude Code 或 Warp 更顺；工作流已经压在 GitHub / Microsoft，Copilot 管理成本更低；已经把任务分发、评审、云端修复压在 OpenAI 生态，Codex 更自然。这半年 Cursor 的轴在变宽 —— Marketplace 分发用法、SDK 编排 agent、手机接管云端 —— 它在从工作台往平台长；但选型的问题一个字没变：这件活的上下文、执行面和验收面在哪里。不要问谁替代谁。The call: for a primary environment that keeps project context, editing, terminal, and review in one window, Cursor; if work lives mostly in shell and command flow, Claude Code or Warp is smoother; if your workflow is already GitHub / Microsoft, Copilot is lower-friction; if task dispatch, review, and cloud fixes already sit in OpenAI's ecosystem, Codex is natural. Cursor's axis has widened this half-year — a Marketplace to distribute practices, an SDK to orchestrate agents, a phone to steer the cloud — it's growing from workbench toward platform; but the sizing question hasn't changed a word: where do this task's context, execution surface, and acceptance surface live? Don't ask who replaces whom.
— III
那条 Computer Use 线The Computer-Use Line.
Cursor 的 agent 能「用电脑」，但用的是它自己沙箱里那台，不是你的。Cursor's agent can 'use a computer,' but it's the one in its own sandbox, not yours.把线画死：Cursor 的 computer use 在云端 agent 的隔离 VM 里发生，用来点开它自己造的软件、验证它改的东西；企业版的 browser / network 控制也是管这个沙箱。3注 3Note 3Cursor Blog · Agents can now control their own computers（2026-02-24）—— Cursor 的 computer use 发生在云端 agent 自己的隔离 VM / 沙箱里，用来验证它自己造的软件，不是控制你本机登录着的 app。Cursor Blog · Agents can now control their own computers (2026-02-24) — Cursor's computer use happens inside the cloud agent's own isolated VM / sandbox to validate the software it built, not to control the apps you're logged into on your machine.3.8 起自动化拉起的云端 agent 默认就带这个能力、给你出 demo —— 还是同一条线：它点的是自己 VM 里的桌面。它不操作你本机登录着的 app；连 iOS 的 Remote Control 也是反着的 —— 那是你从手机指挥你电脑上的 agent，不是它操控你的手机。真正「替你点别的 GUI app」—— 用你的账号发邮件、在你的浏览器里跨站操作、动你打开着的设计稿 —— 是另一类工具：Claude in Chrome、Operator、Claude 的 computer use。Draw the line hard: Cursor's computer use happens in the cloud agent's isolated VM, to open the software it built and verify what it changed; the enterprise browser / network controls govern that sandbox too.3注 3Note 3Cursor Blog · Agents can now control their own computers（2026-02-24）—— Cursor 的 computer use 发生在云端 agent 自己的隔离 VM / 沙箱里，用来验证它自己造的软件，不是控制你本机登录着的 app。Cursor Blog · Agents can now control their own computers (2026-02-24) — Cursor's computer use happens inside the cloud agent's own isolated VM / sandbox to validate the software it built, not to control the apps you're logged into on your machine. Since 3.8, cloud agents launched by automations carry this by default and hand you demos — same line: it clicks the desktop inside its own VM. It does not operate the apps you're logged into on your machine; even iOS Remote Control points the other way — you steering the agent on your computer from your phone, not it steering your phone. Actually 'clicking your GUI apps for you' — sending mail from your account, operating across sites in your browser, touching the design file you have open — is a different class: Claude in Chrome, Operator, Claude's computer use.所以这台元工具的边界是清楚的：项目里、沙箱里、命令行说得清的活，Cursor 替你做完；要驱动你真实的、登录着的 app，切到那一类工具。诚实补一句：这条线在移动，今天清楚、明天可能模糊，每次引用前重新核对官网。So this meta-tool's boundary is clear: work inside a project, inside a sandbox, expressible by a command line, Cursor finishes for you; to drive your real, logged-in apps, switch to that other class. An honest addition: this line is moving — clear today, maybe blurry tomorrow — re-verify the site before citing.
任务Task
Cursor 能不能做Can Cursor do it?
边界Boundary
改项目文件、跑测试、开 PREdit project files, run tests, open PR
能Yes
需要 Git / 测试 / review 兜底Needs Git / tests / review
Cloud Agent 在 VM 里打开 app 点击验证Cloud Agent opens app in VM and clicks to verify
能Yes
只在它自己的远程环境Only in its own remote environment
用你的浏览器账号发邮件 / 改表单 / 操作后台Use your browser account to send mail / fill forms / operate admin UI
不要写成 Cursor 能力Do not describe as Cursor capability
切到浏览器 agent / Computer Use / 手动Switch to browser agent / Computer Use / manual
不是 Cursor Desktop 的官方通用能力Not a general official Cursor Desktop capability
按账号、权限、审计风险另选工具Pick another tool by account, permission, audit risk
SpaceX / Anysphere 这类商业新闻，只能放在背景里：截至 2026-07-10，SEC 层面仍是 6 月那份 8-K —— SpaceX 与 Anysphere 签署 merger agreement，交易还受成交条件和监管批准约束，预计 2026 Q3 完成。4注 4Note 4Space Exploration Technologies Corp. Form 8-K（2026-06-16）—— SpaceX 与 Anysphere, Inc. 签署 merger agreement，交易仍受成交条件和监管批准约束，预计 2026 Q3 完成；截至 2026-07-10 未见完成交割的后续披露。本文只把它作为市场背景，不写成产品能力。Space Exploration Technologies Corp. Form 8-K (2026-06-16) — SpaceX and Anysphere, Inc. entered a merger agreement, subject to closing conditions and regulatory approvals, expected to close in Q3 2026; as of 2026-07-10 no follow-up closing disclosure sighted. This book treats it only as market context, not product capability.产品侧倒是先看到了合作的痕迹：模型文档把 SpaceXAI 列为供应方，Grok 4.5 写明「由 Cursor 与 SpaceXAI 联合训练」、进了第一方池。2注 2Note 2Cursor Docs · Models & Pricing —— 个人档两套额度池：「第一方模型池」（Auto、Composer 2.5、Grok 4.5 —— Grok 4.5 由 Cursor 与 SpaceXAI 联合训练；included 用量宽得多，但非无限）与 API 池（手动选的前沿模型按该模型 API 价）；超出转 on-demand，官方明言「质量与速度不降级」；Max 按 token 的模型 API 价计费（旧 request 制套餐加 20%）；长上下文加价按模型而异；付费个人档 Tab 补全不限量。截至 2026-07-10。Cursor Docs · Models & Pricing — two pools on individual plans: the First-party models pool (Auto, Composer 2.5, Grok 4.5 — the latter jointly trained by Cursor and SpaceXAI; significantly more included usage, but not unlimited) and the API pool (hand-picked frontier models at each model's API rate); past the allowance you go on-demand, with requests 'never downgraded in quality or speed'; Max bills per token at the model's API rate (legacy request-based plans add 20%); long-context surcharges vary by model; Tab completions are unlimited on paid individual plans. As of 2026-07-10.这说明 Cursor 已经进入基础设施级竞争，不说明今天的 Cursor 多了某个功能，也不说明你该因为收购新闻升级。Business news like SpaceX / Anysphere belongs only in background: as of 2026-07-10, the SEC record is still June's 8-K — SpaceX and Anysphere entered a merger agreement, subject to closing conditions and regulatory approvals, expected in Q3 2026.4注 4Note 4Space Exploration Technologies Corp. Form 8-K（2026-06-16）—— SpaceX 与 Anysphere, Inc. 签署 merger agreement，交易仍受成交条件和监管批准约束，预计 2026 Q3 完成；截至 2026-07-10 未见完成交割的后续披露。本文只把它作为市场背景，不写成产品能力。Space Exploration Technologies Corp. Form 8-K (2026-06-16) — SpaceX and Anysphere, Inc. entered a merger agreement, subject to closing conditions and regulatory approvals, expected to close in Q3 2026; as of 2026-07-10 no follow-up closing disclosure sighted. This book treats it only as market context, not product capability. The product side shows the partnership first: the model docs list SpaceXAI as a provider, and Grok 4.5 is stated to be 'jointly trained by Cursor and SpaceXAI,' inside the First-party pool.2注 2Note 2Cursor Docs · Models & Pricing —— 个人档两套额度池：「第一方模型池」（Auto、Composer 2.5、Grok 4.5 —— Grok 4.5 由 Cursor 与 SpaceXAI 联合训练；included 用量宽得多，但非无限）与 API 池（手动选的前沿模型按该模型 API 价）；超出转 on-demand，官方明言「质量与速度不降级」；Max 按 token 的模型 API 价计费（旧 request 制套餐加 20%）；长上下文加价按模型而异；付费个人档 Tab 补全不限量。截至 2026-07-10。Cursor Docs · Models & Pricing — two pools on individual plans: the First-party models pool (Auto, Composer 2.5, Grok 4.5 — the latter jointly trained by Cursor and SpaceXAI; significantly more included usage, but not unlimited) and the API pool (hand-picked frontier models at each model's API rate); past the allowance you go on-demand, with requests 'never downgraded in quality or speed'; Max bills per token at the model's API rate (legacy request-based plans add 20%); long-context surcharges vary by model; Tab completions are unlimited on paid individual plans. As of 2026-07-10. That says Cursor has entered infrastructure-scale competition; it does not mean today's Cursor gained a product feature, nor that you should upgrade because of M&A news.
提示词PromptSubscription self-check
Cursor 订阅自测：
1. 我上个月最常撞的墙是：<额度 / 模型 / Cloud Agents / 团队管控 / 没撞>
2. 我每周 Agent 任务数约：<n>
3. 我是否经常手动选前沿模型或 Max：<是/否>
4. 我是否有 review 多个并行 agent 的产能：<是/否>
5. 我是否需要 SSO / SCIM / 审计 / 模型管控：<是/否>
结论：<Hobby / Pro / Pro+ / Ultra / Teams / Enterprise / 不升级>，理由：<一句话>。Cursor subscription self-check:
1. The wall I hit last month was: <usage / model / Cloud Agents / team controls / none>
2. Weekly Agent tasks: <n>
3. Do I often choose frontier models or Max manually: <yes/no>
4. Can I review multiple parallel agents: <yes/no>
5. Do I need SSO / SCIM / audit / model controls: <yes/no>
Conclusion: <Hobby / Pro / Pro+ / Ultra / Teams / Enterprise / no upgrade>, reason: <one sentence>.
Cursor 订阅自测：
1. 我上个月最常撞的墙是：额度（每周三四次转 on-demand）
2. 我每周 Agent 任务数约：25
3. 我是否经常手动选前沿模型或 Max：否，基本让 Auto 跑
4. 我是否有 review 多个并行 agent 的产能：否，一次只盯一个
5. 我是否需要 SSO / SCIM / 审计 / 模型管控：否，个人单干
结论：Pro+，理由：天天用 Agent 又总被额度打断，但没到要并行多 agent 的程度。Cursor subscription self-check:
1. The wall I hit last month was: usage (went on-demand three or four times a week)
2. Weekly Agent tasks: 25
3. Do I often choose frontier models or Max manually: no, mostly let Auto run
4. Can I review multiple parallel agents: no, I watch one at a time
5. Do I need SSO / SCIM / audit / model controls: no, solo individual
Conclusion: Pro+, reason: I use Agent daily and keep hitting the usage wall, but not enough to need parallel agents.
把这章用到你自己的订阅和工作流上：Apply this chapter to your own subscription and workflow:
写下你上个月撞的墙：额度、模型，还是根本没撞。Write down last month's wall: quota, model, or none at all.
判断你是 Auto 为主还是前沿模型为主 —— 据此看 Pro 的池够不够，还是要 Pro+ / Ultra。Decide whether you're Auto-mostly or frontier-mostly — and whether Pro's pool covers you or you need Pro+ / Ultra.
决定你是 Cursor 单干、还是 + Warp（终端侧）、还是 + 一个能点你真实 app 的工具。Decide between Cursor alone, + Warp (terminal-side), or + a tool that can click your real apps.
确认你没在为一档你从没撞到其上限的额度付钱。Confirm you're not paying for a tier whose ceiling you never hit.
贵的不是高档，
是买了用不上的额度.
The costly tier isn't the high one,
it's the headroom you never hit.
Aklman Library
— 讨论Discussion
讨论Discussion.
评论区初始化中…Initializing comments…
01 / 01
没有匹配结果No matches.
换个关键词，或按 Esc 回到页面Try another keyword, or press Esc to return