Hi, there.
3 Pro-Moves to Save Your Credits
| Action | Impact |
|---|---|
| New Chat for New Tasks | High: Stops the “snowball” effect of token costs in long threads. |
Use @file specifically |
Medium: Instead of letting the AI guess, tell it exactly which file to look at to avoid unnecessary repo scanning. |
| Switch to Gemini 3 Flash | Extreme: Use Flash for “grunt work” (writing docstrings, basic CSS, boilerplate). It is nearly 20x cheaper than Pro. |
_______________________________________________________________________________________
I saw this message in my agent chat.
So, I had a conversation with Gemini.
I thought that it might help the new users to understand better the pricing scale of the Google Antigravity.
I am just going to show you my conversation with Gemini. I hope it will help you to plan your budget better and to save some money. We are talking about the Gemini 3.1 Pro model in this conversation.
Here you go:
________________________________________
Gemini:
With a large multi-folder repository, you are likely hitting the “High Context” pricing tier ($4.00 per 1M tokens) more often than you think. Because Antigravity automatically indexes your folders to provide better answers, it “sees” much more than just the code you’re looking at.
Your Realistic Monthly Expectation
Instead of 2,000 extra prompts, you should budget for about 500 to 800 high-context prompts from those 1,000 credits.
Here is why your “burn rate” will be faster:
-
Repository Map: When you ask a general question (e.g., “Where is the auth logic?”), Antigravity sends a “map” of your entire folder structure. This is cheap but adds up.
-
Import Resolution: If you ask about a function in
main.js, the AI often pulls in the contents of three other files it imports. Your “50-line prompt” suddenly becomes a 20,000-token request. -
The “Long Thread” Trap: If you keep one chat open all day while working on the repo, the model re-reads the entire previous conversation + the code context every time you hit Enter.
Me:
So, I have around 700 prompts per month, correct?
Gemini:
If you only count your 1,000 AI Credits, then yes—at roughly $1.50 per “heavy” prompt (for a large repo), you would have around 600–700 prompts.
However, your actual monthly total is much higher because of your Daily Baseline. Here is the breakdown of how your 2026 Google AI Pro plan works:
Your Total Monthly Capacity
| Source | Limit | Calculation |
|---|---|---|
| Daily Baseline | 100 Pro Prompts / Day | ~3,000 prompts/month (Free) |
| AI Credit Pool | 1,000 Credits / Month | ~700 heavy prompts/month (Paid) |
| TOTAL | ~3,700 prompts per month |
Export to Sheets
Why the “700” number feels right
You are likely focusing on the 700 figure because that represents your “Overage Buffer.” * Your daily 100 prompts are your “sprint” limit.
-
Once those are gone (as in your screenshot), you dip into the credits.
-
If you hit your daily limit every single day, those credits act as a safety net for about 20–25 extra prompts per day.
The “Large Repo” Tax
Because your repository is large, Antigravity uses Context Caching.
-
The Good News: This makes repetitive prompts in the same file much cheaper.
-
The Bad News: If you jump between many different folders, you “bust” the cache, and each prompt could cost 2–3x more ($4.00 per 1M tokens instead of $2.00). In this specific case, your credit pool might only cover 250–300 prompts for the month.
Summary: You have a massive base of 3,000 free prompts, plus a “premium tank” of about 700 extra prompts for when you’re working overtime.
Me:
I pay $20 per month. I have, let’s say, 700 “High Context” prompts for that. So to get to 1000 prompts, which is 300 prompts more, how much on top of my paid $20 will it cost me? Very roughly estimate, please.
Gemini:
To get from 700 to 1,000 prompts (an extra 300 prompts), it will cost you roughly $30 to $60 on top of your $20 subscription.
Since you are working on a large multi-folder repository, here is the breakdown of why the cost falls in that range:
The “Rough” Math
-
Average cost per prompt: ~$0.10 to $0.20.
-
Why? For a large repo, each prompt isn’t just your text; it includes the “context” of your files (tokens).
-
Input: $2.00 per 1M tokens.
-
Output: $12.00 per 1M tokens.
-
-
300 extra prompts: $0.15 (average) × 300 = $45.00.
What changes the price?
-
The $30 Side (Lower Cost): If you use Gemini 3.1 Pro with “Context Caching” enabled. If you stay in the same folder/files, Antigravity caches the code, making subsequent prompts much cheaper (~$0.02 instead of $0.15).
-
The $60 Side (Higher Cost): If your prompts are “heavy”—meaning you frequently ask the AI to look at different parts of the large repo or your context window exceeds 200k tokens. At that point, the price per token doubles ($4.00 per 1M input).
Summary Table
| Total Prompts | Base Cost | Extra Cost (Est.) | Total Monthly Bill |
|---|---|---|---|
| 700 (Current) | $20 | $0 | $20 |
| 1,000 (+300) | $20 | +$45 | ~$65 |
Me:
So, a $20 subscription gives me 700 prompts. Then you are saying that 300 more prompts will cost me an additional $30 - $60?
What are you saying? That the additional prompts cost much more than subscriptional prompts?
Gemini:
I understand why that looks backwards—it feels like the “extra” prompts are significantly more expensive than the ones you get for your $20.
Here is the straightforward reality of how Google is pricing Antigravity in 2026:
The “Bulk Discount” vs. “Retail Price”
Your $20 subscription isn’t actually “buying” a specific number of prompts. It’s a flat-fee “All-You-Can-Eat” pass with a speed limit.
-
The Subscription ($20): You get a daily baseline (approx. 100 prompts/day). If you use them all, you’ve paid about $0.006 per prompt. This is a massive “bulk discount” from Google.
-
The Overages (Pay-As-You-Go): Once you hit that 100-prompt daily wall, Google stops subsidizing you. They switch you to Vertex API pricing, which is the raw “retail” cost of the server power.
Why 300 more prompts cost $30 - $60
Because you are working on a large multi-folder repository, each “prompt” is actually a massive data transfer.
-
Context Loading: When you ask a question, Antigravity sends your code files as “tokens.” In a large repo, a single prompt can easily be 50,000+ tokens.
-
The Math: At the standard rate of $2.00 per 1 million tokens, a “heavy” prompt costs roughly $0.10 to $0.20 in raw server costs.
-
The Result: 300 “heavy” prompts × $0.15 = $45.00.
Is it a rip-off?
It feels like one because the $20 baseline is underpriced to get developers using the tool. Google is likely losing money on your $20 sub if you use all 100 prompts every day. The “Overages” are simply them passing the actual electricity and compute cost directly to you so they don’t lose more money.
Summary Estimate:
-
First 3,000 prompts/mo: Included in your $20 (highly subsidized).
-
Next 300 prompts/mo: Costs roughly $45 (actual market cost).
________________________________________________________________________________________
So, here you go.
It’s time to grab that ice cream from your freezer. 
I hope that this post was helpful for you. 
