Inter-office memo
- To
- Everyone whose Antigravity quota runs out too soon
- From
- Mo, Chief Executive
- Date
- November 16, 2026, 4:42 AM CST
- Re
- Making Antigravity’s quota last
How to make Antigravity’s quota last longer in 2026: 11 ways compared
Start with View Usage in the model picker, then match the model to the task: Google recommends Orchestrator, which finishes “using fewer tokens” than its planner alone. Lower thinking levels, a new conversation per task and fewer swarms do most of the rest, and Pro and Ultra need a free Google Developer Program profile for “higher Google Antigravity quotas.” Enjoy hands routine work to Codex or Claude Code.
Disclosure: Enjoy is our product, and it runs Antigravity. We checked every setting, limit and price below against Google’s, OpenAI’s, Anthropic’s and OpenCode’s own pages on November 16, 2026, and we say where another way is the better pick, including the free ones that don’t need Enjoy.
The ways to make Antigravity’s quota last, compared
Antigravity’s quota comes with your Google account’s plan. Google’s plans page gives accounts without Pro or Ultra “A meaningful quota, refreshed weekly,” Google AI Pro “A high, generous quota, refreshed every five hours until the weekly limit is reached,” and Ultra “The highest, most generous quota, refreshed every five hours.” The model picker counts two groups of models separately, “Gemini Models” and “Claude and GPT models,” each with a weekly and a five-hour limit. These are the ways to make them last, side by side.
| Way | Cost, checked November 16, 2026 | What it cuts | Best for |
|---|---|---|---|
| See where it goes, in View Usage, /usage and /context | Free | Nothing by itself: it shows what’s left in each group, and what fills each turn | Not knowing where your quota went |
| Match the model to the task, and use both groups | Free | A big model’s work on routine steps | Most day-to-day work, on Orchestrator or Hydrogen |
| Lower the thinking level | Free | Reasoning on quick, clear edits | Small, well-scoped changes |
| Go easy on /swarm, /boost and Dynamic Workflows | Free | Extra agents, each working on your quota | Pro and Ultra users of the multi-agent commands |
| A new conversation for each task, and /btw for asides | Free | History carried into every turn | Sessions that run all day or wander |
| Trim what loads with every turn | Free | Rules and MCP tool schemas | Long rule files and many MCP servers |
| Scope the task, and stop it when it drifts | Free | Searching, wrong turns and redone work | Vague or wide changes |
| A Google Developer Program profile | Free, for Google AI Pro and Ultra | Nothing: Google says it’s required for “higher Google Antigravity quotas” | Pro and Ultra subscribers without one |
| A Gemini API key, for scripts and CI | Gemini 3.8 Flash and 4 Hydrogen free on the API’s free tier; paid from $0.75 in and $3.75 out per million tokens | Runs move off your account’s quota | Headless and CI runs in Antigravity CLI |
| Codex, Claude Code or OpenCode for routine work | Codex on ChatGPT Free; Plus $20 a month; Claude Pro $20; OpenCode’s free models | Work moves off Google’s limits | Work that doesn’t need Antigravity |
| Enjoy | A monthly subscription, free for 7 days, with your phone and one computer (pricing) | Shows each group’s limits, and moves work to another agent mid-conversation | Seeing and spreading usage across agents |
Prices are each company’s US list prices, before tax. The first eight ways are settings and habits in Antigravity and your Google account, and cost nothing extra.
Why Antigravity runs out so fast
Google ties the quota to work, not messages. “Under the hood, the rate limits correlate with the amount of work done by the agent, which can differ from prompt to prompt,” says its plans page. “As a result, you receive more prompts when your tasks are straightforward and the agent can complete the work quickly, while more complex tasks consume more quota.” It publishes no sizes for any plan, and “Usage limits for this service are subject to modification.”
So the same request can cost very different amounts, depending on the model, how hard it thinks, how much it reads and how many agents work on it. Google’s docs don’t put a number on any of those, but they say which cost more: Orchestrator finishes “using fewer tokens than running the planner alone,” and a swarm “uses quota much faster than a single conversation.” The ways below follow what Google’s own pages say.
Google’s own developer forum has many threads about the quota running out, and some describe quota that went with little use, such as a Pro subscriber’s report in March that it “went from full to exhausted within hours without running any agents or heavy tasks.” Antigravity’s docs don’t address those reports, and no habit fixes a counting problem: if yours drains while Antigravity sits idle, report it on the forum.
See where it goes: View Usage, /usage and /context
Best for: anyone who doesn’t know their pace. Start here.
In the Antigravity App, the model picker’s View Usage shows “Gemini Models” and “Claude and GPT models,” each with a “Weekly Limit Remaining” and a “Five Hour Limit Remaining,” as Google’s models page shows it, and its plans page adds “You can view baseline quota usage across models on the Settings page.” In Antigravity CLI, /usage, or /quota, “refreshes your model configuration and quota status from the backend” and lists “Your remaining quota for each model group,” per its docs. Since version 1.1.11, agy -p "/usage" prints the same “without starting an agent turn, spending quota, or leaving a conversation behind,” says the changelog, so a script can check before a big job.
For what fills each turn, the CLI’s /context shows “a category-by-category token breakdown”: your messages, the agent’s replies and reasoning, tool calls, rules, skills, MCP tools and more, per its docs.
Price: free.
Watch for: the bars show what’s left, not what each task cost, and Google publishes no sizes behind the percentages. /context measures the model’s context window, not your quota, and Google doesn’t say how the two relate.
Match the model to the task, and use both groups
Best for: most day-to-day work, which doesn’t need Google’s biggest model on every step.
Google’s models page calls Orchestrator the “recommended default” for “Most day-to-day development, such as feature work, bug fixes, refactoring, and multi-file debugging.” It pairs a planner, Gemini 4 Argon or Carbon, with Gemini 4 Hydrogen, which does “the implementation work,” so “you get the planner’s quality while finishing faster and using fewer tokens than running the planner alone.” Choose it in the model picker, or with /model in the CLI. For “Fast, everyday coding and maintenance tasks where you don’t need a separate planning phase,” such as package upgrades, unit tests and documentation, Google suggests Gemini 4 Hydrogen on its own. Keep Carbon or Argon for “Work where every intermediate step needs maximum reasoning depth.”
The two groups have separate limits: in the CLI’s words, “The models in a group share a weekly limit and a five-hour limit.” On Pro and Ultra, Claude Sonnet 5.5 and Opus 5.5 count against “Claude and GPT models,” not the Gemini group, so giving some work to Claude spreads it across both.
Price: free. Orchestrator (Argon) is on Ultra and on paid Pro, or a Pro trial “with eligible premium devices”; Orchestrator (Carbon) is on Ultra only. Free and Google AI Plus accounts have Gemini 4 Hydrogen and Gemini 3.8 Flash, plus Claude Sonnet and Opus 4.6 and GPT-OSS-120b until November 18. Claude 5.5 on Pro is for “non-trial subscriptions only.”
Watch for: Orchestrator “runs with a fixed configuration, so you can’t choose a thinking level for it.” A change waits for the turn in progress: the agent “continues to use the previously selected reasoning model until it completes its steps for that turn.” And accounts without Pro or Ultra lose their second group soon: the models page marks Claude Sonnet and Opus 4.6 and GPT-OSS-120b “Will be removed on November 18, 2026,” as our memo on the change explains.
Lower the thinking level
Best for: quick, clear edits.
Most Gemini models in Antigravity offer Low, Medium and High thinking, and Gemini 4 Hydrogen “defaults to High and adds a fourth level labeled Extra,” says the models page, which frames the choice as one “to balance response speed for quick edits against deeper multi-step reasoning for complex architecture and debugging tasks.” Choose a level for the model in the App’s model picker. In Antigravity CLI, /effort sets it: low gives “faster responses and lighter reasoning, well suited for quick lookups, syntax checks, and straightforward edits,” and medium “balanced speed and reasoning quality for everyday software engineering tasks,” per its docs.
Price: free.
Watch for: Google doesn’t say how much quota a lower level saves. It says the quota tracks “the amount of work done by the agent,” and reasoning is part of what the CLI counts: /context lists “tokens generated by the model’s replies and reasoning steps.” In the CLI, a new level “persists your choice as the default for future sessions,” so raise it again when a hard task comes along. It doesn’t apply to Orchestrator.
Go easy on /swarm, /boost and Dynamic Workflows
Best for: Pro and Ultra subscribers who use the multi-agent commands.
Every extra agent works on your quota. Google is plainest about Swarm: “Every worker is its own agent, so a swarm uses quota much faster than a single conversation,” say its Swarm docs. /boost “delegates focused workstreams to specialized subagents, and independently verifies solutions across iterative rounds,” per its docs, and /run-workflow has the agent write “a short Python script that coordinates many agents for you,” per the Dynamic Workflows docs. Keep them for the hard or wide tasks Google built them for, and do the rest in one conversation.
When you use them, put the helpers on a lighter model. “By default, workflow agents use your conversation’s model,” but you can “Ask for a lighter tier, such as flash_lite or flash, for fast, high-volume checks across many files.” Custom subagents take a model of inherit, flash or pro in their frontmatter. Google does this itself for browsing: “Browser subagents run on a Gemini Flash model by default to keep UI interactions fast and conserve higher-tier quota,” say its browser docs.
Price: free. Swarm and Dynamic Workflows are “available on Google AI Pro and Google AI Ultra plans,” and Boost “on paid plans.” Our memo on Dynamic Workflows covers who has them.
Watch for: idle swarm workers cost nothing, “using no tokens until a post mentions it.” But a worker that runs out “pauses instead of retrying” and “doesn’t restart on its own when your quota refreshes,” so check the board after a refresh.
Start a new conversation for each task, and ask asides with /btw
Best for: sessions that run all day, or wander from one problem to the next.
A conversation carries its history into every turn. Asking a side question in the main chat “adds extra turns to the conversation trajectory and consumes context window capacity on every subsequent turn,” say Google’s /btw docs, and /btw is the way around it: “an ephemeral, single-turn side query” that knows the conversation but isn’t added to its history. It’s in the Antigravity App, since version 2.1.4, and in the CLI. Between unrelated tasks, Google’s /context docs say to “run /clear (alias /new) when switching to a new task that doesn’t require prior conversation context,” or start a new conversation in the App. Within a long task, Antigravity “automatically compacts older turns into checkpoint summaries.”
Price: free.
Watch for: /btw “Answers without tools,” so it “can’t read files, run commands, or search your workspace.” Google presents these as ways to keep the context window lean, and doesn’t say how much quota each saves.
Trim what loads with every turn
Best for: long rule files and several MCP servers.
Some context rides along on every turn. “Loading every tool definition into the model’s context window on every turn consumes tokens and increases latency, even when those tools aren’t used,” say Google’s MCP docs. So Antigravity loads MCP tools lazily by default, listing their names and reading a tool’s schema only when the agent needs it. Setting "eager": true on a tool costs more, because “the tool’s full schema is sent to the model on every turn,” so Google says to reserve it “for frequently used core tools.” If /context shows MCP tools taking “a large share of your context window,” it says to “open /mcp to disable servers you don’t need for the current project.”
Rules ride along too. Antigravity keeps AGENTS.md and GEMINI.md “continuously active,” and for always-on rules it “injects the full content into the system prompt on every turn,” say its rules docs. They share “a 20,000-token budget,” and Antigravity “truncates any single rule file that exceeds 24,000 bytes.” Move long guides into .agents/rules/ with trigger: model_decision: Antigravity then “injects only the rule’s path and description upfront,” and “The agent reads the full rule on demand when your task matches the description.”
Price: free.
Watch for: over budget, Antigravity “demotes the largest rule files from inline text to lightweight pointers,” so a rule you count on may not be in front of the agent. Keep the ones that matter most short and always on.
Scope the task, and stop it when it drifts
Best for: vague requests and changes across many files.
What you ask for decides how much the agent does, and Google says straightforward tasks get you “more prompts.” Name the files: its CLI best practices say to “Give local agents high-fidelity indicators to narrow down reasoning boundaries and minimize token overhead,” and @ adds a file or folder to your prompt. For a big or unclear change, /plan “introduces an intentional planning phase,” because “Writing code before clarifying edge cases may lead to wasted effort, unnecessary diffs, or regressions.” And when the agent heads the wrong way in the CLI, “press the global escape hatch key Esc immediately to interrupt the turn.”
Price: free.
Watch for: planning is work on the same quota. For a one-line fix, skip /plan and say exactly what you want.
Set up your Google Developer Program profile
Best for: Google AI Pro and Ultra subscribers who’ve never claimed their developer benefits.
Google ties Pro’s and Ultra’s higher Antigravity quotas to a free profile. “The Google Developer Program is a no-cost membership,” says Google’s November 13 announcement, and “An active profile is required for Google AI Pro and Ultra subscribers to access developer benefits like monthly Google Cloud credits, higher Google Antigravity quotas, and Google Skills credits.” To make one, claim your plan’s monthly Cloud credits in Google AI Studio: “If you don’t have a profile, the new workflow in AI Studio automatically creates one when you redeem your credits.” Our memo on the credits walks through it.
Price: free.
Watch for: Antigravity’s own plans page doesn’t mention a profile, and Google hasn’t said how much higher the quota is with one. Google describes it as a benefit for Pro and Ultra subscribers, not free accounts or Google AI Plus.
Run scripts and CI on a Gemini API key
Best for: headless and CI runs in Antigravity CLI that you’d rather keep off your account’s quota.
“Run Antigravity CLI with your own Gemini API key instead of a signed-in Google account,” say Google’s authentication docs. “Model requests go directly to the Gemini API, and the CLI never establishes an account session,” and “This suits headless and CI runs, where no browser is available to complete a sign-in.” Set "modelProvider": "gemini" in ~/.gemini/antigravity-cli/settings.json, export the key as GEMINI_API_KEY, and run agy.
Price: per token, at the Gemini API’s prices. Gemini 3.8 Flash and Gemini 4 Hydrogen are “Free of charge” on the free tier. On the paid tier, Flash is $0.75 per million input tokens and $3.75 per million output “through December 31, 2026,” then $1.50 and $7.50, and Hydrogen is $1.50 and $7.50. Pro includes $10 a month in Google Cloud credits and Ultra $40 or $100, per Google One, which Google says you can “put straight toward Gemini API usage,” after a “$5 minimum prepayment.”
Watch for: “Only setting a GEMINI_API_KEY environment variable on its own has no effect.” It’s Gemini only, in the CLI, and on the free tier your prompts are “Used to improve our products”; on the paid tier, they aren’t.
Hand routine work to Codex, Claude Code or OpenCode
Best for: work that doesn’t need Antigravity, so Google’s quota goes to work that does.
Other labs’ agents have limits of their own. “ChatGPT Work and Codex are included in your ChatGPT Free, Go, Plus, Pro, Business, Edu, or Enterprise plan,” says Codex’s pricing, with “GPT-6 Luna at Standard speed in the desktop app, subject to rollout” on Free and Go. “Claude Code is included in all paid plans,” says Claude’s pricing. And OpenCode, an open-source agent, “comes with a set of free models. No account needed,” per its homepage, and runs local models too. To give them all the same project instructions, see our AGENTS.md vs CLAUDE.md guide.
Price: Codex is free on ChatGPT’s Free plan, and Go is $8 a month, Plus $20 and Pro from $100. Claude Pro is $20 a month, or $17 a month billed yearly, and Max is $100. OpenCode and its free models are free.
Watch for: don’t sign them in with your Antigravity login. Google’s FAQ says “Using third-party software, tools, or services to access Antigravity is a violation of our Terms of Service,” which “can result in suspension or termination of your account.” Claude has limits too, since Claude Code “shares the same usage limits as the rest of your plan.” Each agent starts a conversation of its own, so tell it where things stand. Our guide to Antigravity vs Claude Code compares the two.
Enjoy
Best for: keeping Antigravity’s limits in view beside your other agents, and moving routine work to Codex or Claude Code without starting over.
Enjoy runs Antigravity on your Google account, along with Claude Code, Codex, Grok Build, Muse Code, and local models through OpenCode, on Mac, Windows and Linux, each on its own lab’s plan. For quota, it adds:
- Each group’s limits. The agent settings show “the limits Google shows for each group of models, such as Gemini · Weekly,” with when each resets (how).
- Model and effort per message. The agent settings’ Model and Effort columns change them for your next message, and Antigravity’s levels are “Model default, Low, Medium, and High” (how).
- Another agent, mid-conversation. Switch to Codex or Claude Code for the routine part, and “The new agent gets the conversation so far, including replies from other agents” (how).
- A wait instead of a dead end. When the quota runs out anyway, Enjoy “waits until the limit resets, then tries again, and the agent continues where it left off,” or you select Stop waiting and choose another agent (how). Our guide to what to do when Antigravity hits its quota compares the other ways.
Price: Enjoy is one monthly subscription, and the first seven days are free, with a card to start. It includes your phone and any browser and one computer; each other computer is $9.99 a month, and each teammate you invite needs a subscription of their own. See pricing.
Watch for: Enjoy doesn’t add to Google’s quota or make Antigravity use less: the settings above do that. It runs “the Gemini models Google offers your account,” so for Claude through Google, use Google’s own app, and it uses one Google account in every project, so it can’t move Antigravity to a second account the way it does for Claude Code, Codex and Grok Build. Compacting a conversation from Enjoy and its count of what each conversation used don’t cover Antigravity. Waiting needs “Your computer must be on, with Enjoy open, when the usage resets.” Enjoy isn’t open source, and it’s a monthly subscription on top of your AI plans once the free week ends. If Antigravity is the only agent you use, the free ways above may be all you need.
When to stay with the defaults, or upgrade
The defaults are a fair start. Google recommends Orchestrator for “Most day-to-day development,” and each model comes with its own thinking level. Keep Carbon, Argon, Extra and the multi-agent commands for the work that needs them, and save on the routine work around it.
If you’ve done all this and still run out most weeks, the plan is too small. Google’s plans put Google AI Pro at $19.99 a month, or $199.99 a year, and Ultra at $99.99 or $199.99 a month, sold as “5x” and “20x higher usage limits vs. AI Pro.” Under Antigravity, though, Google promises only “Higher” and “Highest rate limits to agent model,” not five or twenty times Pro’s quota, and per Google One, “Ultra plans are billed monthly only.” Google AI Plus, at $4.99, doesn’t change Antigravity’s quota: the plans page gives everyone “not on Google AI Pro or Ultra plans” the same weekly one. For the odd heavy week, Pro and Ultra “can use purchased AI credits (or any one-time promotional credits) for additional overage usage above the baseline quota,” bought from Google One in amounts such as “2,500, 5,000, or 20,000 AI credits.” Our guide to Google AI Pro vs Ultra for Antigravity helps you choose.
Skip the tools that rotate Google accounts for you. Google’s FAQ says “Wrapping agy inside third-party coding agents or applications, or automating account rotation, is not supported under consumer plans.”
How we compared
We read Google’s Antigravity plans, models, FAQ, authentication, CLI, rules, MCP, Swarm, Boost, Dynamic Workflows, subagent, browser and changelog pages, its Google AI plans, Google One and AI credits pages and its November 13 announcement, the Gemini API’s pricing, OpenAI’s, Anthropic’s and OpenCode’s pricing pages and homepages, and Google’s developer forum on November 16, 2026, and every one is linked below. We quote each company’s own words for its limits, and leave out what it hasn’t documented, including other sites’ estimates of how much each habit saves. Plans and limits change often, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.
16.Questions.
Why does Antigravity use up my quota so fast?
- Because the quota counts work, not prompts. Google says its limits “correlate with the amount of work done by the agent,” so “more complex tasks consume more quota.” Bigger models, higher thinking levels, long conversations and extra agents all add work. Check View Usage in the model picker, or /usage in Antigravity CLI, to see which group is going.
Which Antigravity model uses the least quota?
- Google doesn’t publish a rate for each model. It recommends Orchestrator for most work because it finishes “using fewer tokens than running the planner alone,” suggests Gemini 4 Hydrogen for “Fast, everyday coding and maintenance tasks,” and runs browser subagents on Gemini Flash “to conserve higher-tier quota.”
Do Claude models in Antigravity use my Gemini quota?
- No. Antigravity counts “Gemini Models” and “Claude and GPT models” separately, and “The models in a group share a weekly limit and a five-hour limit.” On Google AI Pro and Ultra, moving some work to Claude Sonnet or Opus 5.5 spreads it across both. Free accounts lose their Claude and GPT models on November 18, 2026.
Does a Google Developer Program profile give me more Antigravity quota?
- On Google AI Pro and Ultra, Google says you need one for it. Its November 13 announcement says “An active profile is required for Google AI Pro and Ultra subscribers to access developer benefits like monthly Google Cloud credits, higher Google Antigravity quotas, and Google Skills credits.” The profile is free, and Google hasn’t said how much higher the quota is.
Does /swarm use more quota?
- Yes, much more. Google says “Every worker is its own agent, so a swarm uses quota much faster than a single conversation.” Boost and Dynamic Workflows run several agents too, so ask for a lighter tier, such as flash, for the helpers.
Can I switch Google accounts to get more Antigravity quota?
- Not with tools that do it for you. Google’s FAQ says “automating account rotation, is not supported under consumer plans,” and that using third-party software with your Antigravity login “can result in suspension or termination of your account.”