What Is Antigravity?Antigravity 是什么？
Antigravity (antigravity.google) is Google's agent-first IDE — a fork of VS Code redesigned around autonomous coding agents rather than inline autocomplete. Instead of typing code and getting suggestions, you hand whole workstreams to agents that plan, edit files, run terminal commands, and report back through a mission-control view.Antigravity（antigravity.google）是 Google 的 agent 优先 IDE——一个基于 VS Code 分叉、围绕自主编码智能体重新设计的编辑器。它不是你写代码它给建议，而是把整条任务流交给智能体：规划、改文件、跑终端命令，最后在任务控制台里向你汇报。
It launched in November 2025 with famously generous free quotas — and then tightened them twice: a quiet cut in December 2025, an AI credit system in March 2026, and a full pricing restructure at Google I/O in May 2026 (Antigravity 2.0). As of October 2026, Antigravity has no subscription of its own: the free Individual tier ($0, weekly-refreshing baseline quota) and the paid tiers Pro ($19.99/month), Ultra 5x ($99.99/month, ~5× Pro capacity), and Ultra 20x ($199.99/month, ~20× Pro, cut from $249.99) are Google AI plans that also cover Gemini and Google One. Pro/Ultra quotas refresh every 5 hours up to a weekly cap, with optional pay-as-you-go AI credits for overages.它 2025 年 11 月上线时免费额度非常宽松，之后收紧过两次：2025 年 12 月悄悄砍了一刀，2026 年 3 月上线 AI 积分体系，2026 年 5 月 Google I/O 彻底重组定价（Antigravity 2.0）。截至 2026 年 10 月，Antigravity 本身没有订阅：免费的 Individual 版（$0，基础额度每周刷新）与付费档 Pro（$19.99/月）、Ultra 5x（$99.99/月，约 5 倍 Pro 额度）、Ultra 20x（$199.99/月，约 20 倍，从 $249.99 下调）都是 Google AI 套餐，顺带覆盖 Gemini 和 Google One。Pro/Ultra 额度每 5 小时刷新一次、直至每周上限，超额可按需购买 AI 积分。
The model selector as of October 2026 includes Gemini 3.x (3.1 Pro and Flash models), Claude Opus/Sonnet 4.6 and 5.5 (the 5.5 models are reserved for paying Pro/Ultra subscribers; the 4.6 models are scheduled for removal on November 2, 2026), and OpenAI's gpt-oss-120b. Note: Antigravity does not let you bring your own API key to buy extra quota. Confirm current prices and model availability on the official site.截至 2026 年 10 月，模型选择器包含 Gemini 3.x（3.1 Pro 与 Flash 模型）、Claude Opus/Sonnet 4.6 与 5.5（5.5 仅限付费的 Pro/Ultra 档；4.6 计划于 2026 年 11 月 2 日下线），以及 OpenAI 的 gpt-oss-120b。注意：Antigravity 不支持自带 API key 来扩额度。当前价格与模型清单以官网为准。
Antigravity is Google's admission that the IDE wars are over and the agent wars have begun: the interesting unit is no longer a completion, it's a delegation. The free tier is genuinely useful as a daily driver for casual work, but the product's real bet is bundling — turning "an IDE with agents" into "Google AI plans with an IDE attached". For developers, the honest test is a week of real multi-stream work: if you end up orchestrating parallel subagents instead of babysitting one, it earned its place. Watch the model list though: Google has already removed models (Claude 4.6 out November 2, 2026), and your workflow is only as stable as the selector.
Antigravity 等于 Google 承认：IDE 战争打完了，现在打的是智能体战争。有趣的单位不再是「补全」，而是「委派」。免费档做日常轻度开发是真的够用，但产品的真正赌注是打包——把「带智能体的 IDE」变成「附带 IDE 的 Google AI 套餐」。对开发者来说，老实测试方法是用它干一周真实的多线任务：如果你最后是在指挥一群并行子智能体而不是盯着一个，那就是它赢了。但也要盯紧模型清单：Google 已经开始下架模型（Claude 4.6 在 2026-11-02 下线），你的工作流只和选择器一样稳定。
— 中国吧 AI Tools Hub Editorial Team
Who Should Use Antigravity?谁适合使用Antigravity？
✓ Good Fit For适合以下场景
- Developers who delegate multi-step coding work to agents rather than writing every line习惯把多步编码工作委派给智能体的开发者，而不是每行都自己写
- Anyone already paying for Google AI Pro/Ultra — Antigravity capacity comes bundled已经在付 Google AI Pro/Ultra 的人——Antigravity 额度直接打包在里头
- Teams wanting parallel subagents and background scheduled tasks需要并行子智能体和后台定时任务的团队
✕ Not Ideal For不适合以下场景
- Pure autocomplete devotees — Cursor's Tab engine is still the best in that lane纯粹的补全党——行内补全这条赛道 Cursor 的 Tab 引擎仍是最好的
- BYOK workflows — you cannot bring your own API key for extra quota自带 API key 的工作流——不能用自有 key 扩额度
Getting Started with AntigravityAntigravity 快速开始
1. Download the Antigravity app from antigravity.google and sign in with your Google account从 antigravity.google 下载桌面应用，用 Google 账号登录2. Start on the free Individual plan: check your weekly baseline quota on the pricing page first从免费 Individual 版开始：先到定价页确认本周的基础额度3. Open a project, describe the goal to an agent, and watch the mission-control view coordinate subagents打开一个项目，向智能体描述目标，在任务控制台里看它协调子智能体干活Key Features核心功能
-
Agent-First IDE (VS Code Fork)Agent 优先的 IDE（VS Code 分叉） — Familiar editor surface, redesigned around delegating whole workstreams to agents — not just inline autocomplete.熟悉的编辑器界面，围绕「把整条任务委派给智能体」重新设计——不只是行内补全。
-
Parallel Subagents + Mission Control并行子智能体 + 任务控制台 — Run multiple agents on different workstreams at once and watch them from one mission-control view.多个智能体同时跑不同任务流，在一个任务控制台里统一指挥。
-
Background Scheduled Tasks后台定时任务 — Schedule agents to run in the background (e.g. nightly refactors) without keeping the editor open.把任务排进后台定时跑（比如每晚自动重构），不用一直开着编辑器。
-
Multi-Model Selector多模型切换 — Switch between Gemini 3.x, Claude Opus/Sonnet 4.6/5.5, and gpt-oss-120b per task; the latest Claude models need a paid plan.按任务切换 Gemini 3.x、Claude Opus/Sonnet 4.6/5.5、gpt-oss-120b；最新的 Claude 模型需要付费档。
-
Turbo & Strict ModesTurbo 与严格模式 — Turbo mode lets agents run terminal commands without approval; strict mode requires approval and sandboxes network access.Turbo 模式允许智能体免审批跑终端命令；严格模式要求审批并沙箱化网络访问。
Pros & Cons优缺点
✓ Good Fit For适合以下场景
- True agentic IDE: parallel subagents and background tasks beat single-agent rivals真正的智能体 IDE：并行子智能体与后台任务超过单智能体对手
- Free tier is genuinely usable — full product access with a weekly quota免费档是真能用——完整产品+每周额度，不是残血试玩
- Capacity bundles into Google AI plans you may already pay for额度打包进你可能已经在付的 Google AI 套餐
- Multi-model choice: Gemini 3.x, Claude 4.6/5.5, gpt-oss-120b in one selector多模型可选：Gemini 3.x、Claude 4.6/5.5、gpt-oss-120b 同一选择器
✕ Not Ideal For不适合以下场景
- Quotas have been tightened and restructured twice since launch — no fixed numeric guarantee上线后额度被收紧、重组过两次，没有固定的数字承诺
- No BYOK — you cannot bring your own API key for extra quota不支持自带 key——不能用自己的 API key 扩额度
- Models get removed (Claude 4.6 out Nov 2, 2026) — workflows tied to one model can break模型会被下架（Claude 4.6 在 2026-11-02 下线）——绑死单一模型的工作流可能被打断
- Google's training-use policy for code/prompts is not spelled out — no written no-training commitment on consumer plans代码/提示词是否用于训练没有明确承诺——消费版没有书面「不用于训练」保证
Use Cases应用场景
Popular ways people use Antigravity.人们使用Antigravity的常见场景：
🚀 Multi-Stream Refactors多线重构
Hand a "migrate this service, update tests, rewrite docs" bundle to parallel subagents instead of doing it sequentially.把"迁移服务+更新测试+重写文档"这样的一揽子任务交给并行子智能体，不用一个一个顺序啃。
🌙 Overnight Agent Runs夜间无人值守任务
Schedule background tasks — dependency upgrades, lint sweeps, doc generation — to run while you sleep.把依赖升级、lint 扫描、文档生成排进后台定时任务，睡一觉起来看结果。
🧪 Model Comparison Builds模型对比构建
Run the same coding task across Gemini 3.x, Claude, and gpt-oss-120b from the same selector to see which fits your codebase.同一个选择器里用 Gemini 3.x、Claude、gpt-oss-120b 跑同一任务，对比哪个更吃得下你的代码库。
📦 Delegate-and-Review Workflows委派+审查工作流
Strict mode for untrusted repos: agents ask before acting and run inside a sandbox with restricted network access.不信任的仓库用严格模式：智能体先问后干，在沙箱里跑，网络访问受限。
Similar AI Tools相似 AI 工具
If Antigravity doesn't fit your needs, here are other popular AI tools you might consider:如果Antigravity不符合你的需求，可以看看这些热门 AI 工具：