Google publishes no absolute number for Antigravity usage limits at any tier. Its own plans documentation describes the free allowance as a "meaningful quota", Google AI Pro as a "high, generous quota" and Google AI Ultra as "the highest, most generous quota". Free-tier quota refreshes weekly. Pro and Ultra refresh every five hours until you reach a weekly ceiling. Quota is consumed by how much work the agent does, not by how many requests you send.

That is the complete honest answer, and it is shorter than most people expect. If you came here hoping to find out how many prompts you get, nobody outside Google knows, because Google has never said. What follows is everything that is published, what it means in practice, and how to work out your own ceiling when the vendor will not give you one.

What are Google Antigravity's rate limits?

Antigravity has three quota tiers, not four. The plans documentation groups everyone who is not on Google AI Pro or Google AI Ultra into a single baseline tier, which means Google AI Plus subscribers sit with free users as far as agent quota is concerned. If you upgraded to AI Plus expecting more agent runs, you bought storage.

Tier	What Google calls the quota	Refresh	Overage available?
Free / baseline	"Meaningful quota"	Weekly	No
Google AI Pro	"High, generous quota"	Every 5 hours, up to a weekly limit	Yes, via AI credits
Google AI Ultra	"The highest, most generous quota"	Every 5 hours	Yes, via AI credits

The only figures Google has ever attached to these tiers are the 5X and 20X multipliers it announced for the two Ultra plans at I/O 2026. Both are measured against Pro, and Pro has no published value of its own. The ladder is one, five and twenty times a quantity nobody can see.

How is Antigravity quota actually consumed?

By work done, not by requests sent. This is the single most useful thing to understand about the system and the thing that surprises people most. A simple edit that the agent finishes in one pass costs far less than a long autonomous run that reads twenty files, writes a migration and reruns your test suite three times. Google states this directly: rate limits correlate with the amount of work the agent performs, and you will get many more prompts out of your allowance if your tasks are straightforward.

Two practical consequences follow. First, counting your prompts tells you nothing about how close you are to the ceiling. Second, two developers on identical plans doing what looks like the same amount of work can land in completely different places, because one of them is asking for small changes and the other is handing over whole features.

What does the Antigravity free tier include?

More than most paid tiers elsewhere, which is why the limits question comes up so often. The $0 Individual plan lists Gemini 3.1 Pro, Gemini 3.5 Flash and Gemini 3 Flash alongside Claude Sonnet 4.6, Claude Opus 4.6 and gpt-oss-120b. It also states unlimited Tab completions and unlimited Command requests. The words "basic weekly rate limits" carry all the restriction, and they apply to agent work rather than to autocomplete.

So the free tier is genuinely unlimited in the two places most tools meter you, and rationed in the one place the product is actually differentiated. Reaching Opus-class models inside a paid editor normally means a subscription plus a third-party model allowance you can burn through. Here it costs nothing until the agent quota runs out.

Why does my Antigravity quota show a multi-day lockout instead of a five-hour reset?

Because the five-hour refresh is not the only limit in play. Pro and Ultra refresh every five hours until you reach a weekly ceiling, and once you are against that weekly ceiling the five-hour cycle stops helping you. Users have reported exactly this on Google's own AI Developers Forum: a paid account showing a lockout measured in days rather than the five hours the documentation led them to expect. Both behaviors are consistent with what Google published. The weekly cap is just far less prominent than the five-hour one.

If you are on Pro and seeing multi-day waits, you have most likely hit the weekly ceiling rather than found a bug. The available remedies are to buy AI credits for overage, to move more work to the models that consume less quota, or to wait.

Can you buy more Antigravity quota?

On Pro and Ultra, yes. Google AI Pro and Ultra subscribers can purchase AI credits that are consumed at standard Gemini Enterprise Agent Platform consumption pricing once the baseline quota is exhausted. Free-tier users cannot; when the weekly allowance is gone, it is gone until the week turns over.

For organizations the picture is different again. Antigravity has no seat price. Companies come in through Google Cloud, under Google Cloud terms, with consumption-based Gemini Enterprise Agent Platform billing tied to a Cloud project, and Google notes it is included in selected Gemini Enterprise app subscriptions. That is a meaningfully different purchase from a per-seat license: the spend is variable by design, it lands on an existing cloud invoice rather than a separate software line, and it is worth having a way to watch what your cloud and SaaS spend is doing month to month before you route a team's agent usage through it.

How do you budget for a tool with no published quota?

You cannot forecast it from the documentation, so measure it instead. The method that works is boring and reliable:

Run one real week, not a demo. Give the agent the kind of work it will actually do, at the volume it will actually do it, on the tier you are considering.
Record when you hit a wall, not how many prompts you sent. The useful data point is the day and hour you were throttled and what you had been doing in the hours before.
Repeat on the tier above only if you hit the wall. If a week of honest work never touches the ceiling, the cheaper tier is your answer and no amount of published documentation would have told you that.
Price the failure case. Decide in advance what it costs your team when someone is locked out on a Thursday afternoon. If the answer is "nothing much", the unpublished quota is a non-issue. If the answer is a missed release, buy the predictable tool.

That fourth step is the one people skip, and it is the one that decides the question. An unpublished quota is only a problem in proportion to what being throttled costs you.

How do Antigravity's limits compare to Cursor's?

They are not the same kind of limit, which is why comparing them directly is harder than it looks. Cursor states included model usage in dollars at every tier: $20 on Pro, $70 on Pro Plus, $400 on Ultra, then on-demand at API prices past that. You can check any of those against last month's invoice. Google states quota in adjectives and throttles you rather than billing you.

The honest summary is that Cursor lets you convert usage into a number before you commit and Google does not, while Google gives away at $0 roughly what Cursor charges $20 a month to reach. Which of those matters more depends entirely on whether you are spending your own money or someone else's. We put both side by side, tier by tier, on Antigravity vs Cursor, and the wider entity page covers the rest of the product line on Google Antigravity.

Will the free tier stay this generous?

Nothing in Google's documentation commits to it. The rate limits have already been adjusted more than once since the November 2025 launch, including the move that put free users on a weekly refresh cycle, and Google has been open that demand drove those changes. The current free tier is excellent and worth using today. Building a team's daily workflow on top of it, without a plan for the week the terms change, is a risk worth naming out loud rather than discovering later.

That is the general shape of the trade with metered tools: the allowance is set by the vendor and can move. If you would rather the cost of an agent be a number you set once, an AI coding agent billed at a flat rate takes the quota question off the table entirely, and AI coding tool pricing lays out which vendors meter and which do not.

Keep reading

Antigravity usage limits and rate limits: what Google actually publishes for the free tier, Pro and Ultra