Gemini doesn’t have one price. It has at least four, depending on which door you walk through. The consumer app in your browser is free until it isn’t. The API charges by the token in fractions of a cent. Google Workspace folds it into a per-seat subscription you may already be paying. And then there’s the Ultra tier, which costs more per month than most people’s electricity.
None of that is hidden, exactly. It’s just spread across a pricing page, a help centre article, and a Workspace plan comparison table. Here’s what each bill actually looks like.
The four bills wearing the same name
Before the numbers, get the structure straight. Google sells Gemini in roughly these shapes:
- The free Gemini app — web and mobile, no card required.
- Google AI Pro — $19.99/month for the consumer upgrade.
- Google AI Ultra — $249.99/month for the heavy end.
- The Gemini API — pay-as-you-go per million tokens, for developers.
There are also Workspace plans that include Gemini, which I’ll get to. The key point: paying for one does not get you the other. A Pro subscription gives you consumer features, not API credits you can spend building something.
What the free tier genuinely covers
The free Gemini app has quietly become good enough for most casual use. You get Gemini 2.5 Flash with generous limits, a capped number of Gemini 2.5 Pro queries per day, file uploads, image generation, and Deep Research in small doses. For drafting emails, summarising a PDF, or troubleshooting a spreadsheet formula, it’s plenty.
The API has a free tier too, with rate limits low enough that it’s really a testing sandbox rather than a production option. One caveat worth knowing: on free-tier API usage, Google may use your prompts to improve its models. If you’re pasting client data, that matters more than the price saving.
Where free starts to pinch is when you hit the daily Pro cap mid-task. The app doesn’t gracefully downgrade. It tells you to come back later, usually at the exact moment you’re deep in something complicated.
Google AI Pro at $19.99: the line most people land on
This is the tier Google actually markets, and it packs in more than the chat window. For the monthly fee you get higher limits on Gemini 2.5 Pro, Deep Research, video generation with Veo at a limited daily quota, NotebookLM Plus, and 2TB of Google One storage. It also unlocks Gemini inside Gmail, Docs, Sheets, and Meet for personal accounts.
The 2TB of storage is the underrated part. If you’re already paying about $10/month for Google One storage, the effective cost of the AI features drops to roughly a tenner. That’s a much easier decision than $19.99 from a standing start.
Who should skip it? Anyone whose Gemini use is a handful of quick questions a week. The free tier covers that comfortably, and you can compare how Gemini and ChatGPT split the work on everyday tasks before committing to either subscription.
Google AI Ultra at $249.99: who that’s actually for
Ultra exists for a narrow group: people generating a lot of video, running Deep Think on genuinely hard problems, or needing the highest quotas across the board. It throws in 30TB of storage and early access to features before they reach Pro.
For a writer or analyst, the gap between Pro and Ultra is hard to justify at twelve times the price. The realistic buyer is someone billing video work to clients, where three or four extra Veo generations a week covers the subscription on its own. If that isn’t you, stay on Pro and revisit when your usage genuinely breaks the ceiling.
Gemini API pricing: where the real numbers live
Developers pay per million tokens, split into input (what you send) and output (what comes back). Output costs more, often four to eight times more, because generating text is the expensive part. At the time of writing, the headline rates look roughly like this:
The core model rates
- Gemini 2.5 Pro — $1.25 per million input tokens, $10 per million output tokens.
- Gemini 2.5 Flash — $0.30 input, $2.50 output.
- Gemini 2.5 Flash-Lite — $0.10 input, $0.40 output.
To make that concrete: 1,000 support conversations averaging 500 input tokens and 700 output tokens each comes to about $1.90 on Flash and around $7.60 on Pro. Same workload, four times the bill. The decision isn’t “which model is better” but “which tasks actually need the better model.” Classification, extraction, and reformatting rarely do.
The 200,000-token cliff
Gemini 2.5 Pro has a long context window, and it charges more once you cross 200,000 input tokens. Above that line, input doubles to $2.50 and output jumps to $15 per million. Paste a 400-page document pack into a prompt and you’re on the expensive side of the cliff without noticing.
Chunking is usually the cheaper answer. So is context caching, which lets you store a large shared document and reuse it across many requests at a reduced rate rather than re-sending it each time.
Costs that don’t show up in the headline rate
Three things inflate invoices in ways people don’t anticipate:
- Thinking tokens. Reasoning models generate internal tokens before answering, and you’re billed for them as output.
- Search grounding. Connecting Gemini to Google Search for fresher answers is billed per grounded prompt, in the region of $35 per 1,000 requests, and that stacks on top of normal token costs.
- Image generation. Roughly four cents per image on the current Flash Image model. A hundred iterations while you’re refining a design is $4 from a standing start.
The Batch API cuts costs by about half for anything that can wait, which covers more use cases than teams assume.
Gemini inside Workspace: the per-seat maths
If your company already runs Google Workspace, Gemini likely arrives through the Business Standard or Business Plus plans rather than a separate subscription. Those sit around $14 and $22 per user per month, and Gemini is bundled in. Business Starter, at roughly $7, has much more limited access.
The trap is buying a personal Google AI Pro subscription for the same work, which duplicates a benefit the company already pays for. Before anyone expenses that $19.99, check what the Workspace plan already includes. There’s a real difference between what Gemini in Workspace replaces and the work you’ll still do yourself, and it’s worth understanding the boundary before you budget for it.
Five ways people overpay without realising
- Running every request through the Pro model, including the trivial ones.
- Re-sending the same 50,000-token style guide on every call instead of caching it.
- Leaving thinking budgets at maximum for tasks that need no reasoning.
- Paying for Ultra for the video quota, then not using the other features at all.
- Buying a Pro subscription for someone whose employer already has Workspace with Gemini included.
The pattern in almost all of these is paying for capability you never invoke. AI subscriptions reward habitual use and punish occasional use, so audit your actual usage once a quarter rather than setting and forgetting.
Getting more out of a cheaper tier
The cheapest way to cut a Gemini bill isn’t a discount, it’s fewer turns. A vague prompt that takes six rounds of correction costs more than a precise one that takes two, and the difference compounds across thousands of requests.
That’s true on the API and equally true on a flat-rate subscription, where your limit is queries per day rather than pounds per token. Learning a handful of prompt structures that reliably get there in one shot measurably extends a capped tier. Similarly, if your subscription exists mainly to produce words, it’s worth reading up on where Gemini for writing earns its keep so you’re not burning quota on things you’d write faster yourself.
Picking a tier without overthinking it
Free if you ask a few questions a week. Google AI Pro if you use it most days and could use the storage anyway. Ultra only if video generation or Deep Think pays for itself. For developers, build on Flash, promote individual tasks to Pro, cache anything you send twice, and batch anything that can wait overnight.
One last thing worth doing before you commit: run the same awkward, real-world task through Gemini and a competitor, and count the turns it takes to get something usable. A model that needs four corrections at $0.30 per million tokens can cost more in your time than one that nails it once. That’s the comparison worth making, and the gap between Gemini and Claude on real prompts is a useful sanity check on what you’re actually buying. Google also adjusts API rates and plan inclusions every few months, so check the official pricing page before signing anything off.

