What are AI credits in Antigravity, and how much do they cost?

Updated 2026-07-151,900 searches/moRanked #186 of 519· AI explained
Short answer

AI credits are Antigravity's usage currency, introduced in March 2026 to replace the original quota system. Reported top-up pricing is $25 for 2,500 credits ($0.01 each) or $199 for 20,000. Plans run Free, AI Pro $20/mo, AI Ultra $100/mo, Ultra Max $200/mo. The complaint: Google doesn't disclose the credit-to-token conversion rate.

Why — the first-principles explanation

To understand why credits exist — and why users revolted over them in March 2026 — you have to see the problem they solve for Google.

AI coding agents have wildly variable costs per action. Autocompleting a line might cost a fraction of a cent. Asking an agent to refactor a codebase can burn through millions of tokens across dozens of tool calls, costing dollars. If you charge a flat $20/month, your heaviest users can cost you $500 while your lightest cost you nothing, and you have no way to bill the difference. Antigravity originally launched in November 2025 on a quota system — a fixed number of requests per period. But a quota can't distinguish a one-line completion from a full-repo refactor. Both consume one request. That's economically incoherent for the vendor, and it rewards exactly the usage pattern that costs them most.

Credits fix that by making the unit elastic. Instead of counting requests, you count credits, and expensive operations simply cost more credits. Now the meter tracks something closer to actual cost. In March 2026 Google made that switch, and reported pricing is $25 for 2,500 credits — one cent per credit — or $199 for 20,000, which works out slightly cheaper in bulk.

Here's the part that caused the backlash, and it's the thing you should actually understand. Google does not disclose the credit-to-token conversion rate. So you can know exactly what a credit costs in dollars and still have no idea what a credit buys. That's a one-way information asymmetry: Google can adjust how many credits an operation consumes without changing a single published price, and your bill moves while the price list stays identical. Users protested in March 2026 as the effective price "floated upward." That's the mechanism — nothing on the pricing page had to change for costs to rise.

The subscription tiers sit on top: Free, AI Pro at $20/month, AI Ultra at $100/month, and AI Ultra Max at $200/month — with Ultra Max reportedly dropping from $249.99 after Google I/O 2026. Each includes some credit allowance; credits above that are bought as top-ups.

The practical implication is simple and worth acting on: you cannot forecast your spend from the published price. You can only measure it. Run a typical week, watch the burn, then extrapolate. Any estimate you make before that is guessing.

An example that makes it click

Imagine a taxi that charges in tokens instead of miles. Tokens cost a dollar each — that part's posted clearly on the window. But the driver won't tell you how many tokens per mile. You buy twenty tokens, take your usual trip across town, and it costs six. Fine.

Next month the same trip costs nine tokens. The price of a token never changed — still a dollar, still posted on the window, no announcement, nothing to complain about. You just get less ride. That's the complaint about Antigravity credits: the price is transparent, and the thing the price buys is not.

How to do it

  1. Start on the Free tier and do real work — not a test. The only way to know your burn rate is to measure it, since the credit-to-token rate isn't published.
  2. Watch your credit consumption per task type. Note the difference between autocomplete, a single-file edit, and a multi-file agentic refactor. The last category is where credits vanish.
  3. Extrapolate from a measured week, not from the price list. Multiply your observed weekly burn by four and compare against tier allowances before upgrading.
  4. Pick the tier that matches measured burn: Free for light use, AI Pro at $20/mo for regular individual work, AI Ultra at $100/mo for heavy agentic use, Ultra Max at $200/mo for the highest volume (as of 2026-07).
  5. Buy top-ups only after exhausting the included allowance — reported at $25 for 2,500 credits or $199 for 20,000, with bulk marginally cheaper per credit.
  6. Re-measure periodically. Because the conversion rate isn't disclosed, the credits an operation consumes can change without any published price change. Your last measurement may be stale.
  7. Verify all figures against Google's official pricing page before committing. Antigravity's pricing has changed materially at least twice — quota to credits in March 2026, and tier repricing after Google I/O 2026.

Key facts

Infographic: What are AI credits in Antigravity, and how much do they cost — short answer and key facts
Visual summary — What are AI credits in Antigravity, and how much do they cost?
▶ The 60-second explainer (script)

What are AI credits in Antigravity, and what do they cost? Credits are the usage currency Google introduced in March 2026, replacing the original quota system. Reported pricing: twenty-five dollars for twenty-five hundred credits — a penny each — or a hundred ninety-nine for twenty thousand. Plans run Free, AI Pro at twenty a month, Ultra at a hundred, Ultra Max at two hundred. Now, why do credits exist? Because AI coding agents have wildly variable costs per action. Autocompleting a line costs a fraction of a cent. Asking an agent to refactor a whole codebase burns millions of tokens across dozens of tool calls — that's dollars. Charge a flat twenty a month and your heaviest user costs you five hundred while your lightest costs nothing. Antigravity originally used quotas — a fixed number of requests. But a quota can't tell a one-line completion from a full-repo refactor. Both are one request. Economically incoherent. Credits fix that: expensive operations just cost more credits. Here's the part that caused the revolt. Google does not disclose the credit-to-token conversion rate. So you know exactly what a credit costs in dollars, and you have no idea what a credit buys. That's one-way asymmetry. Google can adjust how many credits an operation consumes without changing a single published price — your bill moves, the price list doesn't. Users protested in March 2026 as the effective price floated upward. Nothing on the pricing page had to change. The practical takeaway: you can't forecast your spend from the price. You can only measure it. Run a real week, watch the burn, then extrapolate. Anything before that is a guess.

What authoritative sources say

Google Antigravity — Pricingofficial — Google's official Antigravity pricing page is the authoritative source for current plan tiers and credit pricing. source ↗
The Register — Users protest as Google Antigravity price floats upwardmedia — Users protested in March 2026 as Google Antigravity's effective price floated upward following the introduction of AI credits. source ↗
DevClass — Users protest as Google Antigravity price floats upwardmedia — Google Antigravity's shift to credit-based pricing in March 2026 drew user protest over rising effective costs. source ↗

People also ask

How much is one AI credit?

Reported at $0.01 per credit — $25 for 2,500, or $199 for 20,000 in bulk, as of 2026-07. Verify on Google's official pricing page, since Antigravity's pricing has changed at least twice.

How many credits does a task use?

Google doesn't publish the credit-to-token conversion rate, so there's no way to answer this from documentation. You have to measure your own usage empirically.

Why did people complain about the pricing?

Because the undisclosed conversion rate means Google can change how many credits an operation consumes without changing any published price. Users reported the effective cost floating upward in March 2026.

Is there a free tier?

Yes, with significantly reduced limits. It's enough to measure your own burn rate on real work, which is the most useful thing to do before choosing a paid tier.

Which plan should I pick?

Don't pick from the price list — measure first. Run a typical week on Free, observe credit consumption per task type, extrapolate, then match that to Pro ($20), Ultra ($100), or Ultra Max ($200).

Related questions