Google Antigravity is still free — but on November 18, 2026, the free plan gets a lot thinner. Google is removing Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b from the free tier, leaving Gemini 3.8 Flash as the only agent model you can use without paying. If you’ve been coding with Claude inside Antigravity for free, you have eight days to decide what to do about it.
That date matters because it’s not the first one Google announced. The models page originally said November 2. Google moved it to November 18, which gave free users two extra weeks — and, if you’re reading this on November 10, eight days of runway. The removal itself hasn’t changed.
What exactly leaves on November 18
Three models carry the same footnote on Google’s Antigravity models page: “Will be removed on November 18, 2026.”
- Claude Sonnet 4.6 — the workhorse most free users reach for when a task needs real reasoning. This is the model that made Antigravity’s free tier feel like a bargain: a full Claude coding model, inside Google’s agent IDE, for $0.
- Claude Opus 4.6 — the heavier Claude, the one you’d pick for the hardest tasks: big refactors, tricky debugging, architecture decisions.
- GPT-OSS-120b — OpenAI’s open-weight model, the third option on the free menu and the only non-Claude, non-Gemini reasoning model there.
After November 18, a free account — and a Google AI Plus account at $4.99 a month, which shares the free tier’s Antigravity limits — has exactly one agent model left: Gemini 3.8 Flash.
Two things worth noting about that list. First, the removal hits the free plan and Google AI Plus, not Pro and Ultra. Paid subscribers keep their Claude access; it just moves up a generation. Second, the three models aren’t being retired from Antigravity entirely. Claude Sonnet 4.6 and Opus 4.6 are being retired across all plans on November 18 — their replacements, Sonnet 5.5 and Opus 5.5, take over on paid plans. GPT-OSS-120b simply leaves. The free tier doesn’t get the replacements.
And if you signed up for Google AI Pro through a promotional partner deal — the Jio offer that bundles 18 months of Pro is the one developers keep asking about on Google’s own forum — the same wall applies. Google’s forum team confirmed that promotional Pro is provisioned as a partner trial, and Claude Sonnet 5.5, Claude Opus 5.5 and Gemini 4 Argon are available on Google AI Pro “for non-trial subscriptions only.” A promotional Pro account keeps the better Gemini 3.8 Flash quota, but it doesn’t unlock the new Claude models.
The full Antigravity price table, as of November 10
Antigravity has no price of its own. There’s no standalone Antigravity subscription to buy. You pay for a Google AI plan, and the plan decides your quota and your model menu. Here’s what each tier costs in the US, before tax, and what it buys inside Antigravity after November 18:
| Plan | Price/month | Antigravity quota | Agent models after November 18 | When you run out |
|---|---|---|---|---|
| Individuals (free) | $0 | “A meaningful quota, refreshed weekly” | Gemini 3.8 Flash only | Wait for the weekly refresh |
| Google AI Plus | $4.99 | Same as free for Antigravity | Gemini 3.8 Flash only | Wait for the weekly refresh |
| Google AI Pro | $19.99 | “A high, generous quota, refreshed every five hours until the weekly limit is reached” | Gemini 4 Argon, Gemini 3.8 Flash, Claude Sonnet 5.5, Claude Opus 5.5 | Wait, or spend AI credits |
| Google AI Ultra | $99.99 | 5x the Pro plan’s tokens | Same models as Pro | Wait, or spend AI credits |
| Google AI Ultra (top tier) | $199.99 | 20x the Pro plan’s tokens | Same models as Pro | Wait, or spend AI credits |
A few things in that table deserve a second look, because they’re the parts most guides skip.
Google AI Plus is the worst value in the lineup for Antigravity users. It costs $4.99 a month and buys you nothing extra inside the IDE — same models, same quota as free. Plus exists for the Gemini app and extra storage, not for Antigravity. If you’re paying for Plus because you thought it unlocked more coding models, it doesn’t. Cancel it or upgrade to Pro; don’t sit in the middle.
Pro’s quota refreshes every five hours, free’s refreshes weekly. That’s a bigger difference than it sounds. A weekly quota means a heavy Monday can leave you locked out until next week. Pro’s five-hour refresh means you hit a wall, go for lunch, and come back to a topped-up allowance. Google doesn’t publish the actual numbers for either quota — “meaningful” and “high, generous” are the official words — so you can’t compare them precisely. What you can say is that the shape of the limit changes: free punishes bursts, Pro forgives them.
Ultra buys volume, not models. The $99.99 and $199.99 tiers give you the same model menu as Pro — Gemini 4 Argon and Claude 5.5 — with 5x or 20x the tokens. Unless you’re running agents all day, every day, Ultra is paying for headroom you probably won’t use. Pro is where the model menu changes. That’s the tier that matters.
There’s no bring-your-own-key. Google’s plans page is blunt: “There is currently no support” for bring-your-own-key or bring-your-own-endpoint for additional rate limits. You can’t plug your own Anthropic API key into Antigravity to keep Claude after November 18, and Google’s FAQ says using third-party software such as Claude Code to access Antigravity “is a violation of our Terms of Service” and can result in suspension or termination.
What Gemini 3.8 Flash can actually do for you
The honest question for every free user is whether Gemini 3.8 Flash is enough on its own. The short answer: for a lot of everyday coding, yes. For the hard stuff, you’ll feel the difference.
Gemini 3.8 Flash is Google’s fast, low-cost model — the one built for high-volume work. It’s quick, it handles straightforward tasks well, and it’s the model Google puts everywhere: the Gemini app, AI Mode, AI Studio, and Antigravity’s free tier. For writing boilerplate, generating tests, explaining code, fixing simple bugs, and scaffolding new files, it’s genuinely capable. Most developers who try it for routine work won’t notice they’ve lost anything.
The practical test is simple: for the next week, before November 18, switch your default model to Gemini 3.8 Flash and do your normal work. Don’t judge it on one task. Run a full week of real work on it. If you finish the week without once switching back to Claude in frustration, the free tier still serves you. If you switch back twice a day, that’s your answer — and it’s cheaper to learn it now, while Claude is still there to compare against, than on November 19 when it isn’t.
The November 18 checklist: run this before the deadline
Eight days is enough to get your house in order. Here’s the list, in the order I’d do it.
1. Check which model your projects actually use
Open Antigravity and look at the model picker in your active conversations. If your long-running projects are set to Claude Sonnet 4.6 or Opus 4.6, they’ll fall back to Gemini 3.8 Flash on November 18 — mid-conversation, with no warning inside the app beyond the date on the models page. Decide now whether each project stays on Flash or moves to a paid plan. Don’t let the first post-November-18 session be the moment you discover the switch.
2. Finish or checkpoint your Claude-dependent work
If you have a big refactor or a debugging session that genuinely needs Opus-level reasoning, do it this week while Opus 4.6 is still available. Better yet, finish it and commit it. The worst version of this deadline is a half-done refactor on November 19, with the model that understood the plan gone and a new model that has to be re-briefed from scratch.
3. Try Flash for a full week (see above)
This is the cheapest experiment in the whole decision. A week on Flash tells you more than any comparison table. If Flash holds up, you just saved yourself $240 a year.
4. Decide on Pro before November 18, not after
If the Flash week tells you that you need Claude, subscribe to Google AI Pro at $19.99 a month before the deadline, while you can still compare Sonnet 4.6 against Sonnet 5.5 on the same tasks. The comparison matters: 5.5 is the model you’d actually be paying for, and you want to know it’s at least as good at your work as the 4.6 you’ve been using free. Also remember the trial catch — Claude 5.5 needs a paid, non-trial Pro subscription. A free trial of Pro won’t show you the Claude models at all.
5. Audit your other free Claude access
Antigravity isn’t the only place Claude lives for free. If your workflow is really “I need Claude for coding,” weigh the alternatives rather than assuming Pro is the only answer. Claude Code is included in Anthropic’s Claude Pro plan at $20 a month. Kiro’s free plan still includes Claude Sonnet 4.5 within 50 credits. The right choice depends on whether you want Claude inside Google’s IDE or Claude inside Claude’s own tools. Google has been consolidating its coding agents into Antigravity — it shut down Jules, its earlier async coding agent, in October 2026 — so expect Antigravity, not a side tool, to be where future model access lives.
6. If you’re on a promotional Pro plan, check your model menu today
Jio-bundled Pro subscribers and anyone else on a partner-promotional Pro plan: open Antigravity, go to the model picker, and check whether Claude Sonnet 5.5 and Gemini 4 Argon appear. If they don’t, you’re on a trial provisioning and they won’t appear on November 18 either.
What just shipped: Remote Agents and Voice Mode (and who gets them)
The model removal isn’t the only Antigravity news this month. Version 2.29.0 shipped November 3, 2026, and it’s the biggest update to the app in a while. But read the fine print before you get excited: both headline features are paid-plan features.
Remote Agents run your agent sessions in isolated Google-managed cloud virtual machines instead of on your computer. Start a task, close your laptop, and the agent keeps working — tests keep running, pull requests keep forming. You can check on sessions and unblock permission prompts from your phone at antigravity.google.com/remote. They’re rolling out to Google AI Pro and Ultra subscribers on personal Google accounts. Not the free plan, not Google AI Plus, and not Workspace or Google Cloud business accounts.
Voice Mode, powered by Gemini Live, turns a conversation with the agent into a two-way spoken dialogue. You talk, the agent answers in real time, you can interrupt it, and you can approve routine permission prompts by voice. Google offers nine voices and playback speeds from 0.75x to 2x. But there are guardrails: destructive actions — deleting files, running commands with elevated privileges, force-pushing to a Git remote — can’t be approved by voice. And in the app, voice mode works only in local conversations for now, not in cloud conversations or Remote Control sessions.
The same release also redesigned the conversation page (the prompt box moved to the bottom, with a greeting at the top, Gemini-app style), redesigned Automations, added a native Jupyter notebook viewer, and added downloads and browsing history to the built-in browser.
One warning about Remote Agents that the launch posts mostly buried: unless you change it, the permission preset is always-proceed (turbo), so remote agents run commands without asking for approval. That’s the point of a cloud agent that works while you’re away — but it’s also a setting you should understand before you hand a remote machine your repository.
Antigravity vs Cursor: the same week, the same shrink
Antigravity isn’t the only free agent tool getting thinner this month. OpenAI is stopping its supply of models to Cursor on November 12 — six days before Antigravity’s Claude removal. Two of the most popular free agent-coding setups are losing models in the same week, for the same underlying reason: the model makers are tightening who gets their frontier models for free.
| Antigravity | Cursor | |
|---|---|---|
| Free tier after this month | Gemini 3.8 Flash only (from Nov 18) | Grok and Composer with limited usage; GPT leaving Nov 12 |
| Cheapest way to keep Claude | Google AI Pro, $19.99/month (flat, in-quota) | Premium plan from $30/month, Claude at API list price on top |
| Claude models | Sonnet 5.5 and Opus 5.5 included in Pro quota | Opus, Sonnet and Haiku 5.5 from Premium, metered |
| Runs on | Mac, Windows, Linux, Googlebook | Mac, Windows, Linux |
| Cloud agents | Remote Agents on Pro/Ultra | Cloud agents from Premium ($30/month) |
The key difference is how Claude is billed. On Google AI Pro, Claude Sonnet 5.5 and Opus 5.5 sit inside your flat $19.99 quota — you don’t pay per token. On Cursor, Claude costs API list price on top of your plan. If you use Claude heavily, Antigravity’s Pro is the cheaper way to keep it. If you use Claude lightly and want Grok or other models Cursor offers, Cursor’s math can work out better.
Is Google AI Pro worth $19.99 a month?
This is the question the November 18 deadline forces, so let’s answer it with real numbers instead of vibes.
Pro costs $19.99 a month, or about $240 a year. For that you get:
- Claude Sonnet 5.5 and Opus 5.5 inside the quota — no per-token billing, no API meter running. Anthropic’s own Claude Pro plan is $20 a month and includes Claude Code, its own agent tool. So the Claude part of Google’s Pro is priced almost identically to buying Claude directly — a dollar less, in fact.
- Gemini 4 Argon — Google’s flagship reasoning model, generally available since October 13, 2026, with a 1M-token context window. It’s the model for the hardest tasks, and it’s not available on any free tier anywhere. (Heads-up if you’re budgeting around Argon: its introductory pricing ends November 30, and the standard rate applies from December 1. Our earlier piece on Gemini 4 Argon pricing doubling on November 30 has the numbers.)
- A quota that refreshes every five hours instead of weekly, plus AI credits when you run out.
- Remote Agents and Voice Mode, the two features from the 2.29.0 release that free users don’t get.
- Everything else in Google AI Pro — the Gemini app’s higher limits, 2TB of storage, and the rest of the bundle — which you’re already paying for if you subscribed for anything else.
The honest math: if you were going to pay $20 a month for Claude Code anyway, Google AI Pro at $19.99 gives you Claude inside an IDE plus Gemini 4 Argon plus the rest of the Google bundle for a dollar less. That’s a good deal. If you only used Claude occasionally in Antigravity and Flash covers 90% of your work, it’s $240 a year for the last 10% — and you should try the Flash-only week first and see if that 10% is real or imagined.
The bottom line
Google Antigravity’s free plan isn’t disappearing. It’s narrowing. On November 18, 2026, Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b leave, and Gemini 3.8 Flash becomes the only agent model on free and Google AI Plus accounts. The price of every plan stays the same — $0, $4.99, $19.99, $99.99, $199.99 — but the value of each one shifts. Free becomes a single-model tool. Pro becomes the only place with Claude and Gemini 4 Argon, plus the cloud agents and voice mode that shipped this month.
Eight days is plenty of time to make the call. Switch to Flash for the week, see how it feels, checkpoint anything that needs Claude, and decide with evidence instead of panic. If Flash is enough, you lose nothing and keep a free tool. If it isn’t, $19.99 a month is a fair price for what Pro now includes — just make sure you’re buying the paid Pro, not a trial or a promotional plan that doesn’t include the models you’re paying for.
Antigravity started as Google’s bet that an agent-first IDE could be free and still good. After November 18, it’s still free — just honest about what free means now.
Frequently Asked Questions
Is Google Antigravity still free?
Yes. The Individuals plan costs $0 and needs no subscription. What changes on November 18, 2026 is the model menu: Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b leave the free plan, and Gemini 3.8 Flash becomes the only agent model available. Tab completions stay unlimited, and the weekly quota stays in place.
What happens to my projects using Claude Sonnet 4.6 after November 18?
They fall back to Gemini 3.8 Flash. Google marks the three models “Will be removed on November 18, 2026” on its models page, and after that date they no longer appear in the model picker on free and Google AI Plus accounts. Conversations set to those models will run on Flash instead. If a project genuinely needs Claude-level reasoning, finish it before the deadline or move to a paid Pro plan, where Claude Sonnet 5.5 and Opus 5.5 replace the 4.6 models.
Does Google AI Plus ($4.99/month) keep Claude in Antigravity?
No. Google AI Plus shares the free tier’s Antigravity limits — same models, same quota. After November 18, Plus accounts also have only Gemini 3.8 Flash. If you’re paying for Plus mainly for Antigravity, it’s the worst value in the lineup: either stay free or step up to Pro at $19.99 a month, which is where Claude and Gemini 4 Argon actually live.
Can I use my own Anthropic API key to keep Claude in Antigravity?
No. Google’s plans page says there is currently no support for bring-your-own-key or bring-your-own-endpoint in Antigravity. Google’s FAQ also says using third-party software such as Claude Code to access Antigravity violates its Terms of Service and can lead to suspension or termination. The only supported way to keep Claude in Antigravity is a paid Google AI Pro or Ultra plan.
What’s the difference between Claude Sonnet 4.6 and Claude Sonnet 5.5?
Sonnet 5.5 is the newer generation that replaces Sonnet 4.6 across Antigravity’s paid plans on November 18. Google lists Sonnet 5.5 and Opus 5.5 as available on Google AI Pro for non-trial (paid) subscriptions only, plus Ultra and Enterprise. Free users don’t get the 5.5 models — the free tier’s Claude access ends entirely with the 4.6 retirement.
Is Google AI Pro worth $19.99 a month for Antigravity?
It depends on your usage. If you run agents for hours a day, need Claude or Gemini 4 Argon regularly, or want Remote Agents that keep coding after you close your laptop, Pro is fairly priced — Anthropic’s own Claude Pro plan is $20 a month for Claude Code alone. If your work is light and mostly boilerplate, Gemini 3.8 Flash on the free tier probably covers it, and you should test a Flash-only week before paying. One caution: promotional Pro plans (such as partner-bundled offers) don’t include Claude 5.5 or Argon, so check your model menu before assuming you’re covered.