Antigravity Plans, AI Credits & Overages Explained

Antigravity Plans and Prices (2026)

Antigravity has no standalone paid tier of its own for individuals. You use it free, and your Google AI plan on Google One sets how much quota you get. The antigravity.google pricing page shows three cards: a $0/month Individual plan, a Google AI Pro card and a Google AI Ultra card (both list a flexible AI credit pool), plus a business plan through Google Cloud.

Sources: antigravity.google/pricing, antigravity.google/docs/plans, Google One plans page (US prices, other countries differ). Google says usage limits are subject to modification.

Antigravity Starter Quota: What the Free Plan Gets

If your app shows Your plan: Starter or a "Starter quota", you are on the free tier. Google's pricing page describes it as: access to a variety of agent models, unlimited Tab completions, unlimited command requests and basic weekly rate limits. The plans doc adds that accounts without Google AI Pro or Ultra get "a meaningful quota, refreshed weekly" and a weekly rate limit.

Antigravity Model Quota by Plan

Quota is tracked per model group, and which models you can pick depends on the plan. From the Antigravity models page (checked October 30, 2026):

Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b are marked on that page as being removed on November 2, 2026. The page also shows Weekly Limit Remaining and Five Hour Limit Remaining percentages in the model selector, with Claude and GPT models listed as their own group. For Claude-specific limits, also read what to do when Opus runs out.

Antigravity AI Credits: Price, Where to Buy, How They Are Spent

• Who can buy: Google AI Pro and Google AI Ultra members only. Not sold for Google AI Plus, not available for purchase in Japan, and not available if you subscribe through a third party, Pixel Pass or Apple billing (Google One Help).
• Where: one.google.com/ai/credits. Sign in, click Get AI credits, choose an amount (the help page gives 2,500, 5,000 or 20,000 as examples), then Buy.
• Price per pack: shown only after sign-in, and we could not verify it. Earlier versions of this page quoted $24.99 for 2,500 credits; treat that figure as unverified and check the checkout screen.
• Auto-reload: after your first purchase you can turn on auto-reload, which tops the balance up when it falls below a threshold you choose.
• Expiry: credits may expire after a period that is set when you acquire them. Check the AI credits activity page.
• Cost per request: credits are deducted at standard API pricing for the specific model and request complexity, so the amount varies. Google does not publish a credits-per-prompt figure.
• Failures: credits count as spent once they perform an activity. If the AI tool fails, your remaining credits are not affected, though a refund can take a while to reappear.
• Shared across products: credits also work in Gemini and Google Flow, so a balance can be used up elsewhere.

Sources: Purchase Google AI credits and Manage your AI credits with Google One.

AI Credit Overages in Antigravity: What the Toggle Does

The in-app text next to the toggle reads: "When toggled on, Antigravity will use your AI credits to fulfill model requests once you're out of model quota. Antigravity will always use your model quota first before using AI credits."

The official plans doc names the setting AI Credit Overages and gives two options:

• Never: credits are never used automatically. You wait until the baseline quota refreshes. You will see a "Baseline model quota reached" notice.
• Always: as soon as a model's baseline quota is used up, Antigravity spends credits. It switches back to baseline quota when that refreshes.

How to turn it on or off

1. Open Antigravity Settings, or open the model selector under the prompt box.
2. Find the AI Credit Overages setting.
3. Pick Always to allow credit spending, or Never to stop at the quota.
4. Check the balance on the Google One AI credits page.

The setting only works if you are on Pro or Ultra and have a credit balance. If you are on the free plan there is nothing to spend. Credit use is per model: running out of one model's quota does not mean the others are used up.

Antigravity Pro vs Ultra

• Refresh: Both refresh quota every five hours. On Pro that only holds until you reach the weekly limit; Ultra has the highest weekly limits.
• Size: Google says Ultra gives 5x or 20x the usage quota of Pro in Gemini and Antigravity, depending on the Ultra tier. These multipliers are vendor-reported, and Google does not publish the base Pro number.
• Models: Ultra lists the highest limits, prioritized traffic and first access to new models. Pro includes Claude Sonnet 5.5 and Opus 5.5 only on non-trial paid subscriptions.
• Credits: both can buy and use AI credits.

For a full side-by-side, see Antigravity Pro vs Ultra plan comparison.

Antigravity Limits, Unlimited Usage and Family Plans

Does Antigravity have a limit? Is there a token limit?

Yes. Every plan has rate limits, and Google says they are mainly set by available capacity and exist to prevent abuse. They track the amount of work the agent does rather than a token count, so a simple edit uses far less quota than a long multi-file task.

How to get unlimited Antigravity limits

There is no unlimited plan. The supported ways to get more are a higher Google AI plan and AI credits. Google does not support bring-your-own-key or custom enterprise tiers for extra limits, and its FAQ says that using third-party tools to access Antigravity, or automating account rotation, breaks its terms and can get an account suspended. For ways to stretch what you have, see how to save tokens and reduce quota use.

Gemini family group and Antigravity quota

Google One help says AI credits for eligible features are pooled across a family group, and that only the plan manager can buy credits, upgrade or downgrade. It does not say whether baseline Antigravity quota is shared. We could not confirm that, so check the Models selector from a family member's account.

Not seeing the plan you paid for?

Google says some models are paid-only and not included in free trials. If a Pro trial hides Claude Sonnet 5.5 or Opus 5.5, the Antigravity FAQ says to submit a request to switch to a paid Pro plan, or upgrade to Ultra. For eligibility errors, see Antigravity account not eligible.

Earlier Guidance (March 2026)

In March 2026, Google quietly restructured Antigravity's pricing from a simple subscription model to a confusing credit-based system. Free tier quotas were slashed by 92%, Pro users lost their 5-hour refresh cycles, and nobody can figure out what a single credit actually buys. This guide breaks it all down.

What Changed in March 2026

When Antigravity launched in November 2025, Google offered generous quotas as part of a free public preview. The Individual plan was $0/month with substantial daily request limits. Pro subscribers ($20/month) enjoyed a 5-hour refresh cycle on their quota, with a separate weekly cap.

In early March 2026, Google announced it was "evolving Google AI plans to give you more control over what you can build." The reality was a sweeping overhaul:

• AI Credits were introduced as a new currency for Antigravity usage
• Free tier daily requests dropped from 250 to 20 — a 92% reduction
• Pro tier quota refresh changed from every 5 hours to weekly for most models
• Credit purchases became the only way to continue working during lockouts
• No clear documentation on what one credit actually represents in compute terms

How the Credit System Works

The new system layers AI Credits on top of the existing dual-limit quota structure. Here's how the pieces fit together:

Credit Pricing

• 2,500 credits — $24.99 ($0.01 per credit)
• 5,000 credits — $49.99
• 20,000 credits — $199 (best bulk rate)

The "AI Credit Overages" Toggle

In Settings → Models, you'll find a new toggle called "AI Credit Overages" with two options:

• Never: When your included quota runs out, you wait for the next free refresh cycle. No charges. This is the default.
• Always: When your quota runs out, Antigravity automatically dips into your purchased credits to keep you working. Credits are consumed until the next refresh.

Quota Refresh Windows

The baseline quota (included with your subscription) refreshes on different schedules depending on your tier and the model:

• Gemini Flash: Still refreshes every ~5 hours across all tiers
• Gemini Pro / Claude models: Pro users now face weekly refresh windows (previously 5-hour). Ultra users retain more frequent refreshes.
• Free tier: Refresh window is weekly, with a hard cap of 20 daily requests

This means that once a Pro user hits their weekly limit on Gemini Pro or Claude, they're locked out for up to 7 days unless they purchase credits. Previously, a 5-hour wait was enough. We break down exactly how the sprint-and-weekly-cap timers interact in our weekly quota cooldown explainer.

Pro vs Ultra: Full Breakdown

Here's how the three tiers compare after the March 2026 changes. If you are specifically weighing whether the jump from Pro is worth it, our Pro vs Ultra plan comparison runs the cost-per-task math side by side.

Free Tier ($0/month)

• Daily requests: 20 (down from 250)
• Tab completions: Unlimited (basic features remain free)
• Models: Gemini Flash only
• Refresh: Weekly
• Credits: None included, can purchase separately

AI Pro ($20/month)

• Daily requests: ~100 (down from ~500)
• Models: Gemini 3.1 Pro (High/Low), Gemini 3 Flash, Claude Sonnet 4.6, Claude Opus 4.6, GPT-OSS 120B
• Refresh: Gemini Flash refreshes every 5 hours; premium models now refresh weekly
• Weekly cap: Yes — once hit, locked out until weekly reset
• Credits: Some built-in credits included (exact amount undisclosed)

AI Ultra ($100/month) — new mid-tier

Added at Google I/O in May 2026, this tier fills the gap that used to force light power users straight onto the old $249.99 plan. It carries roughly 5x the Pro quotas.

• Quotas: ~5x AI Pro limits across the model lineup
• Models: Same five models as Pro, at higher priority
• Weekly cap: Much higher than Pro, but not removed
• Credits: Larger built-in credit allotment than Pro (exact amount undisclosed)

AI Ultra Max ($200/month) — was $249.99

• Quotas: ~20x AI Pro limits — the highest tier available
• Access: "Consistent, high-volume access to our most complex models"
• Models: Same five models, at highest priority
• Weekly cap: Effectively removed at this tier (the key differentiator from Pro)
• Credits: Largest built-in credit allotment

The 92% Free Tier Quota Cuts

The most dramatic change hit free tier users. Here's the before-and-after:

The free tier went from being genuinely usable for small projects to barely functional. At 20 requests per day, you can't complete most development tasks without hitting the wall. Google positions this as a "preview ending" — developers see it as a bait-and-switch.

For context, one developer on the Google AI Developers Forum documented their usage drop: "Before January I could use over 300 million input / 1-2 million output tokens in a week. This week I hit my weekly rate limits at less than 9 million input / 200 thousand output tokens."

Which Models Cost What

Antigravity supports five LLMs, and they consume credits at very different rates. While Google hasn't published exact credit-per-token rates, community testing has established a rough hierarchy:

Cheapest to Most Expensive

1. Gemini 3 Flash — The budget option. Maintains 5-hour refresh even on Pro. Best for simple code generation, formatting, and quick questions. Lowest credit consumption by far.
2. Gemini 3.1 Pro (Low) — Mid-range. Good balance of capability and cost. Uses the standard Gemini 3.1 Pro model with reduced compute allocation.
3. Gemini 3.1 Pro (High) — Full Gemini 3.1 Pro with maximum compute. Noticeably better for complex reasoning tasks but burns quota faster.
4. GPT-OSS 120B — OpenAI's open-source model. Competitive quality but higher credit consumption than Gemini models.
5. Claude Sonnet 4.6 — Anthropic's model. Strong coding performance, moderate credit usage.
6. Claude Opus 4.6 — The most expensive option. Opus generates thousands of hidden "thinking" tokens before writing code, consuming credits roughly 4x faster than Gemini models. A single complex session can burn hundreds of credits.

How to Manage Credits Wisely

Until Google publishes transparent credit values, here's how to stretch your budget based on community-tested strategies. For a deeper playbook on cutting consumption per task, see our guide to saving tokens and reducing quota usage:

1. Keep "AI Credit Overages" on "Never"

The default setting prevents automatic credit consumption. Only switch to "Always" during critical work sessions where you can actively monitor spending. A runaway Opus thinking loop can drain your balance in minutes.

2. Use Gemini Flash as Your Daily Driver

Flash is dramatically cheaper, maintains 5-hour refresh even on Pro, and handles 80% of everyday coding tasks (formatting, boilerplate, simple refactors). Reserve expensive models for complex architectural work.

3. Monitor with the Quota Dashboard

The quota panel in Settings → Models now shows model-specific consumption. Check it before starting work and after complex sessions. For real-time monitoring, see our Antigravity Cockpit guide.

4. Be Specific with Prompts

"Review the auth flow in /src/auth.js" costs far fewer tokens than "Review my entire project." Broad prompts trigger the agent to scan your whole codebase, burning tokens on context before writing a single line.

5. Use Pre-Built Workflows

Workflows execute faster because the AI follows a script instead of planning from scratch. This reduces token consumption per task significantly.

6. Close the IDE When Not in Use

Minimized Antigravity still runs background indexing, which sends API calls that count against your quota on every file save. Fully quit the IDE (Cmd+Q on Mac, File > Exit on Windows) when stepping away. Use a .antigravityignore file to exclude heavy directories. See our indexing optimization guide.

7. The 40% Rule

Community-tested strategy: stop using premium models when you hit ~40% quota remaining per window. This gives you a buffer against the weekly cap and prevents surprise lockouts that could halt your work mid-project.

The 168-Hour Lockout Issue

One of the most disruptive consequences of the March 2026 credit update is a cluster of quota lockout bugs that leave Pro users unable to work for up to 168 hours (7 full days). Google's official position is that the Hybrid Quota System works as designed — but a flood of verified bug reports tells a more complicated story.

How the Lockout Happens

Under the new Hybrid Quota System, AI Pro users receive a baseline quota that refreshes every 5 hours until a weekly cap is hit. Once the weekly cap is exhausted, the timer switches from a 5-hour window to a full 7-day lockout (168 hours). The problem: the weekly cap is not prominently displayed, and the AI Credits update introduced several failure modes that trigger the 168-hour timer incorrectly:

• Zero-usage lockouts: Multiple developers report hitting the 7-day timer immediately after subscribing, with zero recorded usage. The Google AI Developers Forum has active threads titled “[URGENT] Pro Account 7-Day Lockout Bug – Zero Usage” with dozens of confirmations.
• Timer resets that don't reset: The lockout timer sometimes resets to 168 hours mid-countdown, extending the lockout beyond the original window. Users logging in to check progress find the clock has restarted from scratch.
• Instant weekly cap burn: After the AI Credits rollout, some Pro accounts hit the weekly cap after as little as 20 minutes of intensive work, triggering an immediate 7-day lockout instead of the expected 5-hour pause.
• Cross-model contamination: In some cases, exhausting the Claude quota has incorrectly triggered lockouts on Gemini models, which are supposed to use independent quota pools.

Is It a Bug or a Feature?

The honest answer is: both. The 168-hour cap itself is intentional — Google designed it to push heavy users toward the paid Ultra tiers ($100 and $200/month). But the zero-usage lockouts, timer resets, and instant weekly-cap triggers are genuine bugs introduced when the AI Credits toggle was deployed. The two issues are intertwined, making it difficult for developers to know whether they are hitting a legitimate limit or a broken counter.

For practical workarounds while the bugs are unresolved, see our quota troubleshooting guide and the real-time status tools in our Antigravity Cockpit guide.

Developer Community Response

The pricing changes triggered one of the largest developer backlashes Google has seen in the AI space. The story broke in mainstream tech press on March 12–13, 2026, amplifying what had been building in developer forums for weeks.

Press Coverage: March 12–13, 2026

Both The Register and DevClass ran identical-headline coverage on March 12–13 under the title “Users protest as Google Antigravity price floats upward.” The articles documented the transition from generous preview quotas to the credit-based model, the 92% free-tier cuts, and the 168-hour lockout complaints from the Google AI Developers Forum. The coverage noted the particular frustration of power users — many of whom had restructured development workflows around Antigravity — now facing either a 12.5x price jump to the then-only Ultra plan ($249.99) or multi-day work stoppages. (Google later softened this in May 2026 by splitting Ultra into $100 and $200 tiers; see the pricing section above.)

Key Complaints from the Developer Forum

• Opaque credit values: “We're being asked to buy credits without knowing what they purchase” — Google has not documented what one credit equals in tokens, requests, or compute time.
• Bait-and-switch accusations: Developers who built workflows around the generous preview quotas now face 92% reductions with no migration path.
• Forced upselling: At the time, the $230 gap between Pro ($20) and the only Ultra plan ($249.99) felt designed to force power users into the expensive tier rather than offering middle-ground options. Google eventually addressed this by adding a $100 mid-tier in May 2026.
• Weekly lockouts and bugs: Pro users report 5-7 day lockouts after exhausting weekly limits, with several experiencing zero-usage lockouts and timer resets introduced by the AI Credits update (see the 168-Hour Lockout section above).
• Silent changes: Multiple quota reductions were implemented without announcements, discovered only when developers hit unexpected walls mid-project.

The timing was especially sensitive coming just weeks after Google's OpenClaw ban wave in February 2026, where $250/month Ultra subscribers had their entire Google accounts suspended for using third-party tools. Combined with the pricing changes, many developers are openly questioning Google's commitment to the developer ecosystem.

Is Antigravity Still Worth It?

After the March 2026 changes — 92% quota cuts, 168-hour lockout bugs, and the enormous gap between Pro and Ultra — the honest answer depends on your usage pattern. Here's a plain comparison of the three tools most developers are weighing right now:

Side-by-Side: Pro $20 vs Claude Code Max $100–200 vs Cursor Pro $20

The Honest Verdict

Antigravity Pro is still worth $20/month if: you primarily use Gemini models (Flash maintains its 5-hour refresh), your usage is light-to-moderate, and you are not relying on Claude Opus 4.6 for daily work. The native IDE experience and Gemini access remain genuinely good.

Antigravity Pro is poor value if: you do heavy coding sessions with premium models, you have been hit by the zero-usage lockout bug, or you depend on consistent availability. A single 168-hour lockout can erase a week of productivity, which makes $20/month feel expensive compared to tools that do not impose that risk.

Claude Code Max ($100–200/month) costs 5–10x more but provides unlimited Claude Opus 4.6 access with no hard lockouts. If your workflow is heavily Anthropic-model-centric, it removes the anxiety of the weekly cap entirely. The CLI-only interface is the main trade-off.

Cursor Pro ($20/month) matches Antigravity's price, offers a comparable native IDE experience, and does not impose 7-day lockouts. The lack of native Gemini access is the main gap. For developers who do not specifically need Gemini 3.1 Pro, Cursor Pro is the cleaner $20 spend right now.

Bottom line: the March 2026 changes have narrowed Antigravity's value proposition significantly. It remains a strong choice for Gemini-first workflows, but the 168-hour lockout risk and opaque credit system make it hard to recommend as a sole coding assistant for intensive daily use.

What Are Your Options

If the new pricing doesn't work for your workflow, you have several paths forward:

• Optimize your usage: Follow the credit management tips above. Many developers report staying within Pro limits by using Flash as their default model.
• Stack subscriptions: Combine Google AI Pro with a separate Claude Code or Cursor subscription. Use Antigravity for Gemini tasks and a second tool for overflow work.
• Switch to API keys: Google recommends using Vertex AI or AI Studio API keys if you want to use third-party coding agents with Gemini models. This gives transparent per-token pricing instead of opaque credits.
• Explore alternatives: Claude Code, Cursor, Windsurf, and GitHub Copilot all offer different pricing models. See our full alternatives comparison.

FAQ