Antigravity vs Cursor: I gave both the same brief and paid nothing

Contents

Antigravity vs Cursor: which should you use?

Almost every comparison of these two opens the same way: Antigravity is the free one, Cursor is the $20 one. I gave both the identical brief last week and neither of them cost me anything, because Cursor’s free tier built the whole thing in under a minute.

That is not a technicality. It changes which question you should be asking.

	Verdict
Best free option	Tie, genuinely — both built the same app for $0, and Cursor did it faster
Best at checking its own work	Antigravity — it wrote and ran its own unit tests. Cursor wrote none
Best overall	Antigravity, 4.3 to Cursor’s 4.2, and the tenth of a point is entirely that test suite

How they compare on one identical build

Both tools got the same message, once, with no follow-ups: build a single-page subscription tracker with six numbered requirements, including two deliberate traps. Annual plans had to display a monthly-equivalent figure rather than the number you typed, and cancelled items had to leave the running total while staying visible. Each needs two ideas held at once.

I verified every result by serving the app and using it, not by reading the tool’s summary.

Axis	Google Antigravity	Cursor	Winner
Requirements met	6 of 6	6 of 6	Tie
Time to a working app	~2.5 min plus a gate	Under 1 minute	Cursor
Lines written	2,894 across three files	581 across three	Cursor
Ran unattended	No — 1 approval, 1 permission	Yes	Cursor
Wrote its own tests	Yes, 6 assertions	No	Antigravity
Cost of this build	$0	$0 (free Hobby tier)	Tie
Entry price after the free tier	Free indefinitely	$20/mo, plus metered usage	Antigravity
Alley Rating	4.3	4.2	Antigravity

G

Google Antigravity: strengths and gaps

It tested its own work, which almost nothing does. Twenty seconds in it produced an implementation plan, waited for approval, then built. Afterwards it wrote a test file, ran it under Node, and recorded six assertions covering annual-to-monthly conversion, quarterly and weekly normalisation, the running total, cancellation, and a localStorage round trip. I checked each against the shipped app and they held. Of seven tools I have given this brief, two tested themselves and this was one.

The free tier is real and indefinite. Antigravity’s pricing page contains exactly one number and it is $0. That tier lists Gemini 3.5 Flash, Gemini 3.1 Pro, Gemini 3 Flash, gpt-oss-120b and, unusually for a Google product, Claude Sonnet and Opus.

It will not run unattended. The approval gate is by design, and it means one-prompt-one-app is unavailable no matter how well you write the prompt. My run needed one approval click and one permission grant.

It sprawls. 2,894 lines for six requirements, against Cursor’s 581 for the same result. Unasked, it added a six-currency switcher, a category spend breakdown, a command-palette search, grid and table views, and JSON import and export. On a prototype that is free polish; on something you maintain it is five times the code to read.

What it built beyond the brief, unasked:

I asked for	It also shipped
Name, cost, billing cycle	Category, renewal date, icon picker, notes
Monthly equivalent for annual	Six-currency switcher
A running total	Category spend breakdown, “monthly saved” metric
A cancelled section	Reactivate button
localStorage	JSON import and export
Dark theme	Grid and table views, command-palette search

Nobody publishes what the free allowance is. Google’s own plan comparison prices Flow Music Credits to the exact credit and describes Antigravity agent requests only as “limited”, “expanded” or “higher”. You cannot budget against an adjective.

The reliability complaints are consistent and I did not reproduce them. Google’s own developer forum carries the criticism that Antigravity stops mid-plan or partway through an implementation, which one poster describes as making it hard to trust for serious work. My single greenfield session hit nothing of the kind, so I am reporting it rather than confirming it. It is the most repeated complaint about the tool.

C

Cursor: strengths and gaps

It was the fastest of everything I have tested. Under a minute, start to finished app, with no approval gate and no follow-up prompts. Seven tools have taken this brief and nothing else has beaten a minute.

It was also among the tightest. 581 lines across three files, within four lines of the most restrained result I have recorded. It stored the raw cost and billing cycle and derived the monthly figure at render time, which is the correct model and not the obvious one — freezing the derived value is how you get a total that stops reconciling.

The free tier does more than people think. All of the above ran on Hobby, with no card. That is the fact this comparison usually gets wrong.

It wrote no tests. Nothing verified the result except me. Given that the app was correct, that is a smaller practical problem than it sounds, but it is the one axis where Antigravity is plainly ahead.

The paid model is where it hurts. Above the free tier, Cursor charges a subscription plus metered usage billed in arrears at the model’s API rate. I paid for it for eleven months and it came to $510.70 on a plan advertised at $20 — $340 of subscription and $170.70 of usage, with one month reaching $159.22. Nothing malfunctioned. A meter does what a meter does, and the bill arrives a month after the behaviour that caused it.

How they differ on price

Here is the thing worth taking away, because it is the opposite of the usual framing.

For a discrete, well-specified task, they cost the same: nothing. Both free tiers completed the identical brief. If your use of an AI coding tool looks like “build me this specific thing”, you can run either indefinitely without paying, and the comparison is about behaviour rather than money.

For sustained daily work they diverge sharply, and not on subscription price. Antigravity stays free but stops telling you where the ceiling is. Cursor charges $20 a month and then charges again for what you use, in arrears, uncapped unless you set a spend limit.

	Antigravity	Cursor
Free tier	Free indefinitely	Free Hobby tier, no card
What the free tier excludes	Undefined “basic weekly rate limits”	Frontier models, MCPs, cloud agents
Entry paid price	$4.99 via Google AI Plus	$20 Pro
Metered usage on top	None	Yes, at API rate, in arrears
Published usage allowance	No — adjectives only	Yes, $20 to $400 by tier
What I actually paid	$0	$510.70 over eleven months

Both have a transparency problem and they are mirror images. Antigravity will not tell you how much you get. Cursor tells you exactly what everything costs and then bills you a month late, which is technically complete and practically invisible.

If you want a hard ceiling with no thinking, Antigravity gives you one by being free. If you want a hard ceiling with Cursor, it exists — the spend limit on Pro, Pro+ and Ultra — and it is off until you turn it on.

How they differ on speed and workflow

Cursor finished in under a minute. Antigravity took about two and a half, and roughly a minute of that was me reading a plan and clicking approve.

The difference is shape, not just duration. Cursor reads the brief and starts building. Antigravity reads the brief, writes down what it intends to do, and waits. On a fully specified task like mine, that gate is pure friction and it is exactly why Cursor wins the clock. On an underspecified task it is a checkpoint that catches a wrong plan before it becomes wrong code.

The workflows underneath differ just as much. Cursor is an editor you drive with agents on call; Antigravity is an agent manager you delegate to, with parallel workspaces and a browser the agents can drive themselves. That distinction survives my test intact — it is the one thing the existing coverage gets consistently right.

One caveat that applies to everything above: this was a greenfield single-page app from one prompt. It is the shape that flatters fast agents and says nothing about a repository that already exists. On that ground the weight of independent reporting favours Cursor for reliability, and I have not tested it myself.

The features that actually differ

Speed and line counts are what I could measure. They are not what separates these two in daily use, so here is the architectural difference, which my test barely touches.

Antigravity is an agent manager. Its 2.0 release is built around a command centre for parallel local agents, with project grouping and scheduled messaging, plus a CLI, an IDE and an SDK for building agents on its harness. The pitch is that you run several agents at once on different parts of a problem and supervise rather than type.

Cursor is an editor with agents in it. It is a VS Code fork, which means your extensions, keybindings and settings come with you, and the AI lives inside an environment you already know. The agent is something you invoke; the editor is where you live.

Two consequences follow, and they decide more cases than any benchmark:

	Antigravity	Cursor
You are	The manager	The driver
Parallelism	Multiple agents at once	Primarily one task at a time
Browser	Agents drive a browser themselves to check UI	Embedded browser panel you inspect
Models	Gemini first, plus Claude Sonnet and Opus	Broad: Claude, GPT, Gemini, its own Composer
Migration cost	A new application to learn	Your VS Code setup carries over

The browser row is the one people underrate. Antigravity’s agents can navigate and visually validate a UI change themselves, which is a genuinely different verification story from a panel you look at. It is also the capability its own sandbox blocked during my test, when its attempt to serve the app locally failed on a permission error — so I have seen the design and not the payoff.

Why most comparisons of these two are out of date

This is worth checking before you trust anything you read on this matchup, including this page.

Antigravity launched in November 2025 and shipped 2.0 around May 2026, a release that reorganised the product around the parallel agent manager described above. A large share of the comparison write-ups you will find on this matchup carry a November 2025 date — within days of the original launch, describing a version that has since been substantially rebuilt. Two of the most widely cited were published on 23 and 25 November 2025, four days and six days after Antigravity first shipped.

That is not a criticism of those write-ups; they were accurate when written. It is a warning about compounding. A post from launch week describes 1.0, later posts cite it, and an AI summary synthesises both into a description of a product that no longer quite exists. If a comparison does not say which version it tested, assume 1.0. The tell is easy to check: a write-up describing Antigravity as a single-agent IDE, with no mention of the parallel agent manager, predates 2.0 regardless of what its date says.

Everything on this page was measured on Antigravity 2.0 with Gemini 3.6 Flash (High), on 10 August 2026, and on Cursor’s 2.x agent window with Composer 2.5 Fast. Both on their free tiers. If you are reading this a year from now, apply the same suspicion to it.

The stability question nobody puts in the comparison table

Both of these tools have an ownership story that matters more than a feature row, and in opposite directions.

Cursor’s parent company, Anysphere, is being acquired by SpaceX in a $60 billion all-stock deal announced in June 2026 and expected to close in the third quarter — which is now. It is the largest startup acquisition on record. Cursor’s pricing has already been rewritten once, when it moved from counted requests to dollar-denominated usage in 2025, and that change was contentious enough to still be generating explainers a year later.

Antigravity is a Google product that has already consumed another one. Google retired Gemini CLI on 18 June 2026 and moved personal users to Antigravity CLI, cutting the free allowance from roughly a thousand requests a day to roughly twenty and changing it from open TypeScript to a closed binary. That is what consolidation looked like the last time it happened here.

Neither of those is a reason to avoid either tool. Both are a reason to be sceptical of a twelve-month commitment: one is changing owner, the other has demonstrated it will re-tier a free product when the strategy shifts. The free tiers, conveniently, commit you to nothing.

Who should pick Google Antigravity

Anyone who wants an agent that checks itself. The test suite is the real differentiator and it is not a small one. If you are going to trust generated code, a tool that verifies before handing over is worth a slower clock.

Anyone whose budget is a hard zero. Free indefinitely, with Claude and Gemini models included, is a genuinely unusual offer and the single best way to find out whether agentic development suits you.

People who want to watch the plan before the code. The approval gate is a feature if you have ever let an agent run and regretted it.

Not for you if you need unattended runs, or you are maintaining what gets built, or you need to know your limits in advance.

Who should pick Cursor

Anyone who wants to stay in an editor they drive. This is the fundamental difference and it decides most cases before any benchmark does.

Anyone doing discrete, specifiable jobs. Fastest of seven, tightest but one, and free for exactly that pattern of work.

Anyone on a large existing codebase. My test does not show this, but the independent reports consistently do, and it matches the shape of the two tools.

Not for you if you cannot tolerate a variable bill. Set the spend limit on day one, or repeat my $510.70 with your own money.

The final word

The page-one framing of this matchup is that you are choosing between free and $20. You are not. Both tools built the same app for nothing, and Cursor did it faster, so on the work most people will actually try first there is no price difference at all.

What you are really choosing between is two ways of being wrong. Antigravity writes too much code and checks it. Cursor writes tight code and checks nothing. Antigravity will not tell you what your free allowance is; Cursor will tell you exactly what you spent, a month after you spent it.

Antigravity takes it 4.3 to 4.2, and the margin is that test suite — being the only one of the pair that verified its own work is worth a tenth of a point and arguably more. But it is close enough that the tie-break is temperamental rather than technical. If you want to manage an agent, take Antigravity. If you want to drive an editor, take Cursor.

And whichever you lean toward: both free tiers are real, neither asks for a card, and the brief that produced every number on this page took under three minutes on each. Run your own.