Most comparisons of the Gemini API's two tiers are about rate limits, because that is what the free tier runs out of first. The pricing page draws a different line, and it draws it under every single model: a row that says whether your content is used to improve Google's products. On the Free tier the row says yes. On the Paid tier it says no. Everything else on this page follows from that row.
Gemini's Free tier is free of charge, with free input and output tokens, Google AI Studio access, and limited access to certain models. Its price is printed in the row: free-tier content is used to improve Google's products. The Paid tier is prepaid then pay-as-you-go, and the same row says it is not.
Changing the row is a billing account. The rate-limits page says that to move from the Free tier to a paid tier you set up billing in AI Studio, and that the upgrade from Free to Tier 1 typically takes effect instantly, with later upgrades within 10 minutes. There is no waiting period between "a client's document is training data" and "it is not"; there is a form.
Per million tokens in USD: Gemini 3.8 Flash is $0.75 input and $3.75 output through December 31, 2026, and the page already prints the next price, $1.50 and $7.50 from January 1, 2027. Gemini 2.5 Flash is $0.30 input, Flash-Lite $0.10, and Gemini 2.5 Pro $1.25 input. The Paid tier also brings higher rate limits, context caching, the Batch API and Google's most advanced models.
On a drafting job of a couple of thousand output tokens, the Paid tier costs under a cent on 3.8 Flash. The drafting page and the AI job cost calculator put that next to the other providers.
Paying does not remove the ceiling; it replaces a rate ceiling with a spend ceiling. Tier 1 needs a linked, active billing account and carries a $250 billing tier cap and a spend rate limit of $10 per rolling 10 minutes. Tier 2 needs $100 paid plus 3 days from the first successful payment, and caps at $2,000 with $50 per 10 minutes. Tier 3 needs $1,000 paid plus 30 days, with a $200 spend limit per 10 minutes and a cap the page gives as $20,000 to $100,000 and above. Hitting a spend-based limit returns 429 RESOURCE_EXHAUSTED, and because the window is a rolling 10 minutes, the fix is usually to wait.
Both tiers share the same daily clock. Rate limits are measured in requests per minute, input tokens per minute and requests per day, applied per project rather than per API key, and requests-per-day quotas reset at midnight Pacific time. Experimental and preview models are more restricted on both.
The Free tier is for work that is yours and that you would not mind Google reading: a prototype, a personal script, a test of how the model handles your prompt. The Paid tier is for anything a client gave you, because the row says so and because the Paid tier is also where the Batch API and the higher limits live. The data page reads the same question across five providers, and Gemini is the only one where the answer changes with the tier.
Put the Paid tier's price against a small business's real month. A hundred drafts of 2,000 output tokens each is 200,000 output tokens, which at $3.75 per million is $0.75; add the prompts at $0.75 per million input and the month is about a dollar. A thousand short support exchanges are a similar order of magnitude. The $250 Tier 1 cap is two hundred months of that, and the $10 spend limit per 10 minutes is more than the month's whole bill.
So the Paid tier is not a decision about money for a one-person operation. It is a decision about the row, about the higher rate limits, and about the date: the same dollar of work costs two dollars from January 1, 2027, and the page has already said so.
Open the pricing page, find the model you use, and read the row under it. If any client material will pass through the project, link a billing account now; the page says the upgrade is instant. Set your own spend limit below the $250 cap if you want a softer stop than a 429. And put January 1, 2027 in the calendar, because the Paid tier's headline price doubles that day and the page already says so.