Google Antigravity Review (2026): Pricing, Features & Honest Verdict

Google Antigravity is an agentic development platform: an IDE, a CLI, a Python SDK and editor extensions that run several coding agents in parallel, free to download with no subscription. The tooling is strong and the pricing page publishes exactly one number. Best for: developers who want frontier agents at no cost. Price: $0/mo, paid tiers unpriced. Rating: 7/10.

What is Google Antigravity?

Antigravity is Google’s answer to Cursor and Claude Code, and it is not one product but five. Antigravity 2.0 (v2.11.0 at the time of writing) is a desktop command centre for running multiple local agents at once, grouping conversations into Projects, working across several workspaces and scheduling recurring tasks. The Antigravity CLI (v1.1.22) does the same from a terminal, with headless mode and background subagents. The SDK (v0.1.15) lets you prototype custom agents in Python on top of the same harness. There is a full agentic IDE, and there are official extensions for VS Code, JetBrains, Zed, Xcode and Visual Studio in preview.

It first launched in November 2025 and has shipped hard since. Version 2.0 arrived in May 2026, the CLI two days later, IDE extensions in August, and Remote Control (driving a local session from another device) in late August. Nine blog posts went up between 21 July and 27 August 2026. Downloads cover macOS 12 and later on Apple Silicon or Intel, Windows 10 64-bit and later on x64 or ARM64, and Linux on glibc 2.28 or newer. The homepage’s own summary is that Antigravity is “available at no charge for developers”, and that is the fact everything else in this review orbits.

What are Antigravity’s key features?

Frontier models, including other companies’ models, on the free tier

The documented model table gives free users Gemini 3.7 Flash, 3.6 Flash, 3.5 Flash and 3.1 Pro as reasoning models, plus Claude Sonnet 4.6 (thinking), Claude Opus 4.6 (thinking) and GPT-OSS-120b. You pick per conversation from a dropdown, with reasoning effort set to low, medium or high on the Gemini models. Google shipping Anthropic’s flagship model inside its own free IDE is not what we expected to find, and it is the single strongest reason to try this.

Artifacts instead of a wall of diffs

Rather than dumping changes into a chat window, agents produce reviewable artifacts: an implementation plan, a walkthrough, and screenshots from browser recordings. The browser integration runs in a separate Chrome profile with an allowlist and denylist, so a frontend agent can drive a real browser against your app without touching your logged-in session. For UI work this is the part that separates it from a chat box bolted onto an editor.

Parallel agents with real controls

Subagents, background tasks, headless mode, per-action permissions, hooks, sidecars and skills are all documented. There are slash commands for deep reasoning (/boost), multi-agent teams (/teamwork-preview), code search, diffs, permissions and session resume. MCP is supported, along with plugins, a status-line customiser, Vim mode and voice dictation. The surface area here is closer to a platform than an editor feature.

A usage viewer that shows you a percentage

Run /usage and Antigravity reports weekly limit remaining and five-hour limit remaining, tracked separately for Gemini models and for the Claude and GPT models. That separation is useful: burning through Opus does not cost you Gemini capacity. It is also the only quantitative window you get into your own quota, and it reports a percentage of a number Google never tells you.

How much does Google Antigravity cost?

Antigravity is free to download and use, with no subscription required. The pricing page lists four plans: Individual at $0 a month, Google AI Pro, Google AI Ultra, and an Organization plan through Google Cloud. The Individual tier includes access to the agent models, unlimited Tab completions, unlimited Command requests and “basic weekly rate limits”. Pro adds “more generous rate limits” and a “flexible AI credit pool”; Ultra adds more of both. The Organization plan runs under Google Cloud terms with consumption-based API pricing through the Gemini Enterprise Agent Platform.

Compare that to what the category charges and the free tier looks remarkable. Cursor’s Hobby plan is free with limited agent requests, Pro is $20 a month, and Teams is $40 per user per month. GitHub Copilot runs Free, Pro at $10 a month (which includes $15 of AI credits), Pro+ at $39 (including $70 of credits) and Max at $100 (including $200). Zed is free forever with Pro at $10 a month including $5 of tokens. Antigravity gives away Claude Opus 4.6 access that its nearest paid rivals meter carefully.

Now the part that costs it a point. Four plans are on that pricing page and exactly one price appears on it: $0. Google AI Pro, Google AI Ultra and the Organization plan carry no figure at all.

Why does the pricing page publish only one number?

We check pricing pages against vendor documentation on every review, and Antigravity’s gap is the widest we have logged. Three of the four plan cards carry no price. Every quota on the page is qualitative: “basic weekly rate limits”, “more generous rate limits”, “flexible AI credit pool”. Each of those phrases carries an asterisk, and the only footnote on the page reads “rolling out to select customers”, which is attached to a different row. The asterisks on the rate limits resolve to nothing.

The documentation does not close the gap. The AI Credits page tells you to see “the main Plans page” for “a detailed explanation of baseline quotas, how credits are consumed, and plan eligibility”. That Plans page then says Ultra users get “the highest, most generous quota, refreshed every five hours”; Pro users get “high, generous quota, refreshed every five hours until weekly limit reached”; everyone else gets “meaningful quota, refreshed weekly”. There is not a single number in it. It goes further and says the baseline limits “are primarily determined to the degree we have capacity”, that they correlate with “the amount of work done by the agent, which can differ from prompt to prompt”, and that limits are “subject to modification”. Overage credits are “consumed at standard Gemini Enterprise Agent Platform consumption pricing”, a rate card on a different product’s documentation.

You can read the whole site and not learn how many agent runs any tier buys, what one credit costs, or what Pro and Ultra are priced at. For a free tool that is a shrug. For anyone deciding whether to move a team onto a paid Google AI plan for the higher limits, the one input to that decision is missing on purpose.

Do Antigravity’s own pages agree with each other?

Not consistently, and the mismatches point in opposite directions.

The pricing page says the free tier’s agent models are “Gemini 3.5 Flash, Gemini 3.1 Pro, Gemini 3 Flash, Claude Sonnet and Opus 4.6, gpt-oss-120b”. The documentation’s model availability table gives free users Gemini 3.7 Flash and Gemini 3.6 Flash as well, and does not list “Gemini 3 Flash” at all. Antigravity’s own blog announced Gemini 3.6 Flash on 21 July and 3.7 Flash on 13 August. So the pricing page undersells the free tier by two model generations while advertising one that appears to be retired.

The Plans documentation lists “access to third-party models” as a benefit reserved for Google AI Ultra. The model availability table on the next page over gives Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b to Free, Plus, Pro and Ultra alike. Those two statements cannot both be true.

The same table carries the finding we did not expect: Enterprise is the only column where all three third-party models are marked unavailable. A developer on the free tier gets Claude Opus 4.6. A developer at a company on Gemini Enterprise does not. Paying Google more removes models here rather than adding them, and no page says so in words. There is also a “Google AI Plus” tier in that table which does not appear on the pricing page anywhere.

Two other limits are stated plainly in the docs and belong in any evaluation. Bring-your-own-key and bring-your-own-endpoint are explicitly unsupported, so you cannot buy your way past a rate limit with your own Anthropic or OpenAI account. And the FAQ says Antigravity is “currently available for personal Google accounts in approved geographies”, advising you to try an @gmail.com address “if having challenges with Workspace Google accounts”. A tool marketed to enterprise developers that may not accept a corporate identity is a real deployment blocker, not a footnote.

Who is Google Antigravity best for?

Use Antigravity if you are an individual developer or a small team who wants the current frontier agent stack without a subscription. The free tier is not a trial: it is generally available, it includes Claude Opus 4.6 and Gemini 3.7 Flash, and the artifact review flow and browser agent are better than most paid alternatives at the same job. If your work is frontend, the separate-Chrome-profile browser agent alone is worth the download. It is also a sensible second editor to keep alongside Cursor, because it costs nothing to hold onto.

Skip Antigravity if you need predictable capacity. The limits are undocumented, explicitly capacity-dependent, subject to change without a published figure, and you cannot lift them with your own API key. Skip it if your team lives on Google Workspace accounts and cannot sign in with personal Gmail. Skip it if you are outside the approved country list, if anyone on the team is under 18, or if your security review needs a clear answer about training on your code, because the FAQ’s answer to its own data-collection question is to read the terms of service.

What are the best Google Antigravity alternatives?

Cursor is the incumbent and the most direct comparison. Hobby is free with limited agent requests, Pro is $20 a month, Pro+ and Ultra sell 3x and 20x the agent limits respectively, and Teams is $40 per user per month with SSO, pooled context and team-wide privacy mode. Cursor also documents a privacy mode that guarantees your code is not used for training by Cursor or its model providers, which is the clearest statement on that question in this group.

GitHub Copilot is the cheapest way to buy a defined quota. Free, Pro at $10 a month, Pro+ at $39 and Max at $100, with the AI credit allowance published for each tier ($15, $70 and $200 respectively). You are getting less agent autonomy than Antigravity, and you are getting a number instead of an adjective.

Zed suits people who want speed and a small editor rather than a platform. Free forever, Pro at $10 a month with $5 of tokens included and usage-based billing beyond that, Business at $30 per seat per month. It also lets teams route through an existing AI contract instead of buying credits twice.

Final verdict: is Google Antigravity worth it?

Yes, for the obvious reason: it costs nothing and it is good. Five surfaces across three operating systems, extensions for every major editor, artifact-based review, a browser agent in an isolated profile, subagents, hooks, MCP and a shipping cadence of roughly two features a week. Handing free users Claude Opus 4.6 alongside Gemini 3.7 Flash is a competitive move Cursor and Copilot cannot match at their free tiers. Download it, run it beside whatever you already pay for, and see which one you reach for.

The 7.0 reflects everything you cannot find out. A four-plan pricing page with one price, quotas described only as basic, generous and most generous, a documentation page that promises a detailed explanation of quotas and delivers adjectives, a stale model list contradicting the model table two clicks away, third-party models listed as an Ultra benefit while the free tier already has them, an Enterprise tier that quietly loses them, no bring-your-own-key escape hatch, and sign-in that may reject your work account. None of that matters much at $0 a month. All of it matters the moment you try to standardise a team on it, and that is the decision Google has made hardest to research.