Google AI (Gemini API)

Google sells Gemini through the Gemini Developer API, with a free tier and a paid tier. On 2026-10-31 Gemini 4 Argon costs $2 in and $10 out per million tokens until 30 November 2026, then $4 and $20. Flash models are far cheaper. Free-tier content is used to improve Google products. Paid-tier content is not.

Facts at a glance

Prices checked 2026-10-31. For the complete price list and the Standard, Batch, Flex and Priority tiers, see Gemini API pricing. This entry covers Google as a supplier: what to expect from the terms and the limits, and how agents connect.

Models offered

Google also lists text-to-speech, live translation and transcription models. The two date clauses are the thing to model before you budget. Argon doubles in price on 1 December 2026, and the Flash line doubles on 1 January 2027. Anything you ship this quarter on Argon enjoys the lower price for a month at most. Budget at the later price and treat the early weeks as a discount.

Google states that "every token the model processes is counted and billed, including thinking tokens". Thinking is priced as output, so a reasoning model can bill several times the length of the visible answer. A thinking budget, where a model offers one, caps it. For model details see Gemini models, Gemini Pro and Gemini Flash.

Free tier versus paid tier

This is the section that matters most, because the data terms change with the tier.

The terms define the tiers by billing, not by the product name. Google AI Studio counts as a Paid Service even when it is free of charge, as long as the account has access to a Cloud project with an active Cloud Billing account, or is a Workspace enterprise account. The Gemini API counts as a Paid Service only when you reach it through a Cloud project with an active billing account. Unpaid quota in the API is an Unpaid Service. Nothing private should go through an unbilled project.

Data terms in detail

For Unpaid Services, the terms say Google uses the content you submit, and the responses, to provide, improve and develop Google products, and that human reviewers may read and annotate it. Google advises against submitting sensitive or confidential data on that tier.

For Paid Services, the terms say Google does not use your prompts, including system instructions, cached content and files, or your responses to improve its products. It processes them under its Data Processing Addendum for products where Google is a data processor. It also says Google logs prompts and responses for a limited period of time, solely to detect and prevent violations of the Prohibited Use Policy, to maintain safety and security, and for required legal or regulatory disclosures. That data "may be stored transiently or cached in any country in which Google or its agents maintain facilities."

Read that last sentence as a residency statement. The Gemini Developer API terms do not promise a region, and the log period is described only as "limited". Teams that need a stated region usually look at the Vertex AI route in Google Cloud instead, where regional controls are part of the product. We did not read Vertex's terms for this entry.

The terms also carry regional modifications. If you are in the European Economic Area, Switzerland or the United Kingdom, they require you to use Paid Services when you make API clients available to users in those places.

Rate limits and usage tiers

Google ties limits to a project's usage tier and shows your live limits in AI Studio. The published spend-based limit is evaluated on a rolling 10-minute window.

Tier 2 and 3 qualification counts cumulative spending on Google Cloud services for the linked billing account, not only the Gemini API. If you hit the spend limit the API returns a 429 RESOURCE_EXHAUSTED error, and the docs say to wait and retry, reduce expensive requests, or ask for an increase. Google says the stated limits are not guaranteed and capacity may vary. In rare cases an upgrade request can be denied even if you meet the criteria.

Priority inference holds its own rate limits, with defaults of 0.3 times the standard limit for each model and tier. We did not find a published table of free-tier requests per minute on the pricing page. Check the rate limit view in AI Studio.

Service levels and caching

Google sells the same model at several service levels. Using Argon as the example, with the same date clauses:

The Flex and Priority terms are described only briefly on the pricing page. Argon also has a Contributor tier, a private preview by invitation at $1 in and $5 out, where content is used to improve Google's products. We would not plan around it.

Gemini caches in two ways. Implicit caching is automatic, and it is the only kind Argon supports. Explicit caching adds a storage fee per million tokens per hour on other models: $0.50 for Gemini 3.8 Flash until the end of 2026 and $1.00 after, $1.00 for Flash-Lite and 2.5 Flash, and $4.50 for the Pro models. A cache read costs a tenth of standard input on Flash and Flash-Lite and a twentieth on Argon. The storage fee means explicit caching only pays if you re-read the cache often enough to cover it. A long, repeated prefix in an agent loop pays back fast, and a one-off document does not.

A worked cost example

Take a job with 10 million input tokens and 2 million output tokens. The arithmetic is ours, from the published prices, with no caching and no thinking overhead.

The same job costs exactly twice as much on Argon after 1 December. Batch halves each row. Thinking tokens are extra output on top of the table, so a reasoning-heavy run will cost more than shown. On Flash-Lite the output side is already the larger share of the bill, which is why short answers matter even more on the small model.

Setting up and avoiding the usual snags

1. Create a key in Google AI Studio. The key belongs to a Cloud project. Check that the project has an active billing account if the data matters.
2. Check the tier. A new project starts on Free. Tier 1 needs a linked billing account, which is a few minutes of work. Tier 2 and 3 take days by design.
3. Choose the model by tier. Argon and 3.1 Pro Preview are not on the free tier, so a free-tier key returns an error for them.
4. Handle 429s. The RESOURCE_EXHAUSTED error covers the spend rate limit. Retry with backoff, and cut context size if you hit it repeatedly.
5. Mind long prompts. The Pro models cost more above 200K tokens, while Argon and Flash have one price.
6. Watch audio. Gemini 2.5 Flash charges $1.00 per million input tokens for audio against $0.30 for text, image and video. Flash-Lite 3.5 uses one input price across modalities.

A fair concern is how stable these numbers are. Google changes preview models and promotional prices often. Two of the current prices carry an expiry date by design. Re-read the pricing page before any budget sign-off and treat our table as a snapshot of 2026-10-31.

Which agents can use it, and how

Google's own agent is Gemini CLI, and it takes three kinds of sign-in. The agent's README describes a Google sign-in tied to a Cloud project, a Gemini API key, and Vertex AI. The API-key route gives usage-based billing:

export GEMINI_API_KEY="your-key-from-ai-studio"
gemini

The Vertex route is written for enterprise teams:

export GOOGLE_API_KEY="your-key"
export GOOGLE_GENAI_USE_VERTEXAI=true
gemini

See Gemini CLI and Antigravity for Google's other agent surface. Third-party agents connect with the same key. OpenCode takes a Google provider through /connect. Aider accepts a Gemini model name together with GEMINI_API_KEY. Which privacy tier you land in depends on how the agent signs in, so check whether the key belongs to a project with active billing before pointing a work repository at it. Gemini CLI's data terms vary by sign-in method and are not in its README.

Alternatives

Who should pick Google

A pre-launch checklist

1. Confirm the project has an active Cloud Billing account before sending real data.
2. Budget at the post-promotion price for Argon and Flash.
3. Set a thinking budget on models that offer one.
4. Use implicit caching on Argon. It does not support explicit caching.
5. Pick Batch for jobs nobody is waiting on.
6. Check your live limits in AI Studio rather than any table on the web, including this one.
7. If you need a region, read the Vertex AI terms first.

Our take

Google has the most generous way in. A free tier with real Flash access lets anyone prototype at no cost. The bargain has a clear price in the terms: free-tier content trains and is read by people, and paid-tier content does not. The practical advice is simple and strict. Put a billing account on any project that sees real data. Beyond that, the Developer API is vaguer than Anthropic or OpenAI about retention length and region, and the two scheduled price doubling dates make Argon a worse buy than the launch price implies. For a team that needs stated regions and enterprise paperwork, the Google Cloud route is the better door.

We have not run tests for this page. It reflects Google's published documents on 2026-10-31. Google does not pay for placement here.

Related pages

Gemini API pricing, Gemini models, LLM API prices, Gemini CLI, ChatGPT vs Claude vs Gemini.