Antigravity Quota: 5-Hour vs Weekly Limits, Resets & How to Check

How Often Does Antigravity Quota Reset?

Google's plans doc gives the reset window by plan. These are the vendor's words, and Google says limits can change:

Google does not publish how many prompts or tokens each window holds. It says limits correlate with the amount of work the agent does, so a simple task uses less than a complex one. The docs also do not say whether the weekly window is a rolling seven days from your first use or a fixed day of the week, so read the countdown in the app instead of guessing. Source: antigravity.google/docs/plans.

Weekly Allowance vs 5-Hour Allowance: Which One Wins?

Think of two meters. The five-hour meter refills every five hours. The weekly meter caps the total you can use across the week. The Pro wording, "refreshed every five hours until the weekly limit is reached", means the weekly meter is the ceiling.

So if you have used all of your weekly allowance but still show five-hour allowance left, the weekly limit takes precedence. A Google staff reply on the AI Developers Forum (June 2026) says the same thing: Pro has a five-hour rolling limit, but also a weekly baseline limit, and reaching it triggers a cooldown until the next weekly cycle. Treat this as vendor-reported behavior.

The models page shows both meters per model group, labelled Weekly Limit Remaining and Five Hour Limit Remaining, with Claude and GPT models listed as their own group. Running out in one group does not by itself mean another group is empty.

How to Check Antigravity Usage and Quota

In the app (IDE or desktop)

1. Open the model selector under the prompt box.
2. Read Weekly Limit Remaining and Five Hour Limit Remaining for each model group. The reset time is shown as a short duration, with the full text in a tooltip (changelog).
3. Open Settings to see baseline quota usage across models.

Google's help pages say the Models selector shows your baseline quota consumption, and that the Settings page shows usage across models. Enterprise and Business accounts have no per-model quotas, so the usage panel is removed for them.

In Antigravity CLI

1. Start the CLI and type /usage (alias /quota) in the prompt box.
2. Press Enter. The Model Quotas panel opens and refreshes your quota from the backend.
3. Scroll with the arrow keys or J and K, and close with Esc or Q.

agy -p "/usage" prints the same information. When printed to a terminal, reset times appear in your local time zone. Piped and --json output use UTC timestamps (changelog). Source: CLI /usage docs.

AI credit balance

Credits are tracked separately from quota. Open the AI credits activity page in Google One to see your balance and usage history. Google also notifies you in Antigravity if you do not have enough credits for a request or have run out.

For live monitoring outside the app, see the Antigravity Cockpit guide. Google staff have said there is no dedicated weekly-limit dashboard, so third-party tools are unofficial.

What to Do When You Run Out of Antigravity Quota

1. Read the message. A "Baseline model quota reached" notice means the quota for that model is used up. Check the timer in the model selector to see which refresh you are waiting for.
2. Switch models. Quota is tracked per model group. If Claude is empty, a Gemini model may still have room. On Pro, the Claude Sonnet 5.5 and Opus 5.5 models are limited to paid, non-trial subscriptions. See what to do when Opus runs out.
3. Use AI credits. On Pro or Ultra with a credit balance, set AI Credit Overages to Always. Antigravity spends credits once the baseline quota is gone, and switches back when it refreshes. Never makes it wait instead. Credits are spent at standard API pricing for the model, so watch the balance.
4. Upgrade the plan. Ultra lists the highest limits, at 5x or 20x the Pro usage in Gemini and Antigravity (vendor-reported). Prices are in the plans and credits guide.
5. Spend less per task. Narrow prompts, fewer files in context and a cheaper model for routine edits all stretch quota. See how to save tokens and reduce quota use.

Why Your Antigravity Timer Says Days, Not Hours

If the countdown reads several days on a Pro account, you have most likely used up the weekly limit for that model group, so the timer shows the weekly refresh. Forum users have reported timers of 5 to 7 days, and some were hit after a single prompt. Google staff replies attribute the long timers to the weekly limit. Whether any of those single-prompt cases are bugs is not confirmed by Google in the sources we checked, so report it through the support channels if the usage bar does not match what you did.

If one model suddenly feels weaker or the app swapped models, that is a separate issue covered in model swap and silent downgrade. For error messages, see fix Antigravity quota issues and 503 and high-traffic rate limit fix.

Earlier Guidance (March 2026)

Antigravity's quota system is the single most confusing and contentious part of Google's AI coding IDE. After two major overhauls in March 2026 — one that enraged the community and a partial revert days later — the system now sits in a messy middle ground. This guide explains exactly how the dual-limit system works, what each model costs, how to monitor your usage in real time, and battle-tested strategies to avoid getting locked out mid-project.

How Quotas Actually Work (Antigravity 2.x, 2026)

Antigravity does not count messages or prompts. It counts "Work Done" — a proprietary compute metric that factors in input tokens, output tokens, hidden thinking tokens, and tool-call overhead. Two prompts that look identical to you can consume wildly different amounts of quota depending on how the model processes them internally.

The system operates on two overlapping limits:

1. Sprint Quota (5-hour cooldown): A short-burst allowance — roughly 250 units on Pro — that refreshes every 5 hours. When you exhaust this bucket, you see the familiar "Quota refreshes in X hours, Y minutes" countdown. After the timer expires, your sprint bucket refills — but only if you still have weekly capacity remaining.
2. Weekly Baseline Cap: A larger ceiling — around 2,800 units on Pro — that governs your total usage across an entire week. Each sprint deducts from this weekly pool. Once the weekly cap is exhausted, the 5-hour refresh stops working entirely. You face a multi-day lockout (up to 7 days) until the weekly cycle resets.

Think of it like a phone data plan: the sprint is your "high-speed data" that resets every 5 hours, and the weekly cap is your monthly data ceiling. Once the ceiling is hit, you're throttled regardless of the sprint timer.

One Weekly Pool Across All Four Surfaces

Since the platform split into four surfaces — the Antigravity 2.0 IDE, the agy CLI, the SDK, and the browser IDE — the most common misconception is that each surface hands you its own allowance. It does not. All of them draw from the same weekly "Work Done" pool tied to your account.

In practice this means the agy CLI — the terminal agent that replaced the sunset Gemini CLI on June 18, 2026 — is not a separate bucket you can lean on when the IDE locks out. An agent you leave running in a terminal deducts from the exact same weekly cap as your editor, and running several agents in parallel across projects drains the pool faster, not slower. One Pro user reported an "Individual quota reached" error after just two prompts in agy because an IDE session had already spent most of the week's allocation.

The March 2026 Timeline: What Changed and What Was Reverted

March 2026 was the most turbulent month in Antigravity's history. Here's exactly what happened, in order:

The partial revert on March 13 restored the 5-hour cooldown for Pro users, but did not remove the weekly cap. This means the system is now a hybrid: you get your 5-hour sprints back, but once you burn through your weekly allocation across multiple sprints, you still face a multi-day lockout. The net result is better than the worst-case March 5 version, but far more restrictive than what Pro users had before March 2026.

Per-Model Quota Breakdown

Each model family maintains its own independent quota pool. Exhausting your Claude allowance does not (in theory) affect your Gemini quota. Here's how the five available models compare in terms of quota consumption:

The key takeaway: model choice is your biggest lever for quota management. Switching from Opus to Gemini Flash for routine tasks can extend your weekly quota by 4–6x. For a deeper comparison of model capabilities and when to use each, see our credits and pricing guide.

The 40% Capacity Bug After Cooldown

One of the most reported issues since the March 13 revert is the 40% capacity bug: when a 5-hour cooldown expires and your sprint quota "refreshes," it only restores to approximately 40% of the original capacity instead of refilling completely.

Here's what developers are experiencing:

• First sprint of the week: Full capacity — you can run multiple complex sessions before hitting the sprint limit.
• After first cooldown: Sprint refills to ~80–90% capacity.
• After second cooldown: Sprint refills to ~60% capacity.
• After third cooldown and beyond: Sprint refills to only ~40% capacity and stays there until the weekly reset.

This creates a diminishing returns pattern where each successive sprint window gives you less and less usable quota. By mid-week, Pro users report getting only a handful of Opus prompts per 5-hour window before hitting the cooldown again.

Free vs Pro vs Ultra: How Quotas Differ

The three subscription tiers have dramatically different quota behavior. Here's the complete comparison after all March 2026 changes:

* Ultra's "unlimited" is not truly unlimited — Google reserves the right to throttle at extreme volumes — but in practice, no Ultra user has reported hitting a hard cap.

The $230 gap between Pro and Ultra remains the core frustration. Google is effectively asking power users to pay 12.5x more to avoid the weekly lockout. There is no intermediate tier. For a full pricing analysis, see our Antigravity credits and pricing guide.

How Credits Interact with Cooldowns

The AI Credits system sits as an overlay on top of the quota system. Here's how the two interact:

1. Baseline quota runs first. Your sprint and weekly quotas are consumed before any credits are touched.
2. When baseline depletes: If the "AI Credit Overages" toggle is set to "Always" in Settings → Models, Antigravity automatically switches to deducting from your credit balance. You keep working without interruption.
3. When set to "Never" (default): Once baseline depletes, you're locked out until the next refresh. No credits are consumed. This is the safer setting.
4. Credits and cooldown timers: Purchasing credits does not reset your cooldown timer or refill your baseline quota. Credits are a parallel resource, not a quota refresh.

Credits cost $24.99 per 2,500 ($0.01 each), with bulk rates available at $49.99/5,000 and $199/20,000. Pro users receive 1,000 monthly credits included; Ultra users receive 25,000. Unused credits do not roll over between months. For the full breakdown, see our credits pricing guide.

Best Strategy: Spreading Usage Across the Week

The weekly cap is the real constraint — not the 5-hour sprint. If you front-load all your heavy work on Monday and Tuesday, you'll hit the weekly cap by Wednesday and face a multi-day lockout for the rest of the week. The developers who avoid lockouts consistently follow a simple principle: pace your compute across the full 7-day cycle.

The Pacing Formula

Divide your weekly quota into daily budgets. If you have roughly 5 full sprint windows worth of weekly capacity, that's about 3–4 moderate sprint sessions per day across a 5-day work week, or 2–3 sessions per day if you work 7 days.

The Model Rotation Strategy

Since each model family has independent quota pools, rotating between Gemini and Claude models effectively doubles your available compute. Use Gemini Pro for one task, switch to Claude Sonnet for the next, then drop to Flash for routine work. This distributes drain across multiple pools instead of exhausting one.

What Happens When You Hit the Limit

When your sprint quota depletes, the IDE shows a countdown timer ("Quota refreshes in X hours, Y minutes"). When your weekly cap depletes, the behavior changes:

• All premium models lock simultaneously — Gemini Pro, Claude Sonnet, and Claude Opus all show the same multi-day countdown. In some cases, the cross-model lockout bug also locks Flash.
• The lockout timer shows 6–7 days, counting down to your weekly reset.
• Tab completions continue to work — basic code completion is free and not affected by quota.
• Gemini Flash usually remains available — it maintains its own 5-hour refresh cycle and is unaffected by premium model lockouts in most cases.

Your Fallback Options

1. Switch to Flash: If Flash is still available, it handles 80% of everyday coding tasks (formatting, boilerplate, simple refactors, documentation).
2. Enable AI Credit Overages: Temporarily switch the toggle to "Always" to use purchased credits. Monitor carefully.
3. Don't count on the agy CLI as a bypass: The agy CLI (which replaced the sunset Gemini CLI on June 18, 2026) draws from the same weekly pool as the IDE, so a weekly lockout locks the terminal too. It's useful for Flash-tier work while your sprint recovers — not a way around the weekly cap.
4. Switch to a secondary tool: Developers who stack subscriptions (e.g., Antigravity Pro + Claude Code or Cursor) can seamlessly switch during lockouts.

For server-side errors like 503s that look similar to quota exhaustion, see our 503 and rate limit troubleshooting guide.

How to Check Remaining Quota in Real Time

Antigravity's built-in quota display is unreliable — the bar sometimes shows 60% remaining while you're already locked out. Here are the methods that actually work:

Method 1: Settings → Models Panel

Open Settings → Antigravity User Settings → Models. This panel shows per-model quota bars and the "AI Credits" balance. Check it before starting a session and after complex work.

Method 2: The Cockpit Dashboard

The Antigravity Cockpit provides a more detailed real-time view of quota consumption, including historical usage graphs and per-model breakdowns. It's the most reliable way to see where you stand. See our complete Cockpit setup guide.

Method 3: Fuelcheck Command

The community-built fuelcheck CLI tool queries your Antigravity quota status directly from the terminal. It's open-source, auditable, and does not consume quota itself:

Understanding the "6-Day Refresh" Change

In mid-March, some Pro users noticed their weekly reset timer changed from 7 days to 6 days. This was not announced by Google and was discovered through community monitoring. Here's what we know:

• The 6-day window is not universal. Some Pro accounts show a 6-day cycle, others still show 7 days. The determining factor is unclear — it may be tied to account age, region, or an A/B test.
• The cycle starts from your last full reset, not from when you hit the cap. If your weekly quota reset on Monday at 2pm, the next reset is Sunday at 2pm (6 days) or Monday at 2pm (7 days).
• The practical difference is significant: A 6-day cycle gives you roughly 52 resets per year vs. 52 on a 7-day cycle — but the 6-day version means your reset day shifts backward each week, eventually cycling through all days of the week.
• Partial refills during the wait: Some developers report seeing a small quota restoration at the 3–4 day mark within a 6-day cycle, though this may be related to the 40% capacity bug rather than an intentional mid-cycle refill.

To check your own cycle, note the exact timestamp when your weekly quota resets and compare it to the previous reset. If the gap is consistently ~144 hours (6 days) rather than ~168 hours (7 days), you're on the shorter cycle.

Tips from Power Users Who Make Quotas Work

These strategies come from developers who have consistently avoided the weekly lockout for weeks without upgrading to Ultra:

1. The 40% Rule

Stop using premium models when your sprint bar hits ~40% remaining. This gives you a buffer against unexpected compute spikes (a complex prompt that triggers a thinking loop) and ensures you don't accidentally exhaust the weekly cap mid-session.

2. Flash as Daily Driver, Pro for Decisions

Use Gemini Flash for 80% of your work: formatting, boilerplate, documentation, simple refactors, and quick questions. Switch to Gemini 3.1 Pro or Claude Sonnet only for architectural decisions, complex debugging, and multi-file refactors. Reserve Opus for at most 1–2 critical sessions per week.

3. New Chat After Every Major Task

Hit Cmd+L (New Chat) after completing each major task. Long conversations accumulate context tokens that inflate every subsequent prompt. Resetting the chat window drops the per-prompt cost back to baseline.

4. Be Surgical with Prompts

"Review the auth flow in /src/auth.js" costs a fraction of what "Review my entire project" costs. Broad prompts trigger the agent to scan your whole codebase, burning tokens on context loading before writing a single useful line. Point the model at specific files.

5. Use .antigravityignore

Background indexing sends API calls on every file save. A well-tuned .antigravityignore file can cut background quota consumption significantly:

6. Close the IDE When Away

Minimized Antigravity still runs background indexing, which sends API calls against your quota. Fully quit the IDE (Cmd+Q on Mac, File > Exit on Windows) when stepping away for more than 30 minutes.

7. Use Workflows and Rules

Pre-built Workflows execute faster because the AI follows a script instead of planning from scratch. This reduces token consumption per task. Custom Rules similarly focus the agent's behavior and prevent wasteful codebase scanning. For token-saving techniques in depth, see our complete token-saving guide.

8. Stack Subscriptions Strategically

Many power users combine Antigravity Pro ($20) with Claude Code ($20) or Cursor Pro ($20). This gives you independent quota pools across two tools for $40/month total — far cheaper than Ultra at $250/month, with more total compute available. Use Antigravity for Gemini tasks and the secondary tool for overflow.

9. Deplete Before Sprint Reset

If you're approaching a 5-hour reset with sprint quota remaining, spend it on low-priority tasks (formatting, documentation, test generation). Unused sprint quota doesn't roll over — it's wasted. This doesn't affect the weekly cap, but maximizes the value of each sprint window.

FAQ