Gemini API

What the Gemini API actually meters

Tokens, counted four ways: input, output including thinking tokens, cached token count, and cached token storage duration. Google's billing FAQ lists those four and nothing else.

Two qualifiers matter more than the headline rate. Prompts above 200,000 tokens on Gemini 3.1 Pro Preview move to a second card, where input doubles to $4.00 and output rises to $18.00. Input climbs 100% and output 50%, so the long-context penalty isn't one multiplier. Thinking tokens bill as output at the full rate, so on Flash models reasoning costs five times what the prompt did.

Caching runs implicitly, on by default for Gemini 2.5 and newer. Cache reads bill at 0.1x base input, and a hit needs a 4,096-token prefix on every Gemini 3 model. Storage bills per million tokens per hour, $4.50 on 3.1 Pro Preview and $0.50 on Flash.

Grounding meters per request. Gemini 3.x models share 5,000 free Google Search requests a month, then bill $14 per 1,000. One user prompt can fire several searches, and Google charges for each. Google Maps grounding carries the same allowance and the same $14.

How credits work

Gemini API credits are a prepaid balance, not a usage allowance. Paid accounts default to Prepay: load a minimum of $5 and a maximum of $5,000, and usage draws down in near real time.

The mechanics are strict. Credits expire 12 months after purchase. Refunds don't exist on Prepay except when you switch to Postpay, which closes the account and returns the balance; close it any other way and the balance is forfeited. Deduction order runs promotional Google Cloud credits first, then prepaid funds, but only while the prepaid balance is positive. At $0 the Cloud credits stop being consumed too, so promotional credit alone can't keep an account alive.

Auto-reload tops the balance up at a threshold you set, and a monthly auto-charge limit caps how much it spends per cycle before switching itself off.

What happens when you hit the limit

Everything stops, and Google built four separate stops. A $0 Prepay balance kills every API key on that billing account at once. A project spend cap pauses one project. A billing account tier cap pauses them all until the 1st of the next month, at $250 on Tier 1, $2,000 on Tier 2 and $20,000 to $100,000 on Tier 3. A spend-based rate limit returns 429 on a rolling 10-minute window, at $10, $50 and $200 across those tiers.

None of these are exact, and Google says so. Billing data lags about 10 minutes, so batch jobs and agent sessions can overrun a cap before the system catches up. Free-tier rate limits aren't published as numbers in the docs at all: you're told to read them in AI Studio.