Does the free AI API tier train on your data?

Yes, on most of them. A free AI API tier is an API plan that costs no money and instead grants the provider broader rights over the text you send it. Google's Gemini API terms are the clearest example. Under the unpaid tier, "human reviewers may read, annotate, and process your API input and output," and Google uses that content "to provide, improve, and develop Google products and services and machine learning technologies" (Gemini API Additional Terms of Service).

The same page carries a blunt instruction. "Do not submit sensitive, confidential, or personal information to the Unpaid Services."

Every free-tier roundup ranks these plans by requests per day. Almost none of them read that sentence. It is the one that decides whether you can legally ship the thing you are building.

What you are actually paying with

The trade is data, not money. Providers that charge nothing recover the cost by using your traffic to improve their models. That is a reasonable bargain for a weekend project. It is a different bargain once a real user's text passes through the key.

The distinction that matters is not free versus paid. It is whose data you are spending. Your own prompts are yours to give away. Your users' prompts are not.

Which providers train on free-tier data

The policies diverge sharply, and the split does not follow price. Here is what each provider's own documentation says.

Provider	Trains on your data?	Source
Google Gemini (unpaid tier)	Yes, plus human review	Gemini terms, unpaid services
Google Gemini (paid tier)	No, and prompts are not used to improve products	Gemini terms, paid services
Anthropic API	No, by default, unless you opt in via feedback	Anthropic's privacy centre
OpenAI API	No, with roughly 30-day abuse-monitoring retention	OpenAI's data controls guide
Mistral	Depends on the product's default setting	Mistral's commercial terms
OpenRouter free models	Depends on the downstream provider	OpenRouter's provider logging documentation

Two things stand out.

The first is that Google's free and paid tiers are governed by opposite rules. On the paid tier, Google states it does not use your prompts or responses to improve its products, and logs data only for policy enforcement and legal disclosure. Adding a credit card does not just raise your quota. It changes the contract.

The second is that Anthropic and OpenAI apply the same no-training default to all API keys. Anthropic's policy states plainly: "By default, we will not use your inputs or outputs from our commercial products to train our models." That default does not weaken because you are on a smaller plan.

Mistral's commercial terms sit in the middle. The company says it will not train on customer data except where a product is "set to opt-in by default" and you have not opted out. Which bucket your plan falls into is an account setting, not a headline. Check it before assuming.

Why this is a compliance problem in Europe

Human review is the clause that turns a preference into a legal exposure. GDPR is the European Union's data protection regulation, and it requires a lawful basis for every act of processing personal data (Regulation 2016/679, Article 6). Sending a user's personal data to a third party who may have staff read it is processing you must disclose and justify. You need a lawful basis. You need it in your privacy notice. You likely need it in a data processing agreement.

A free tier gives you none of that paperwork, because the provider has told you in advance not to send that data at all.

This is narrower than the compliance duties arriving under the EU AI Act, which govern how AI systems are built and deployed. Free-tier data use is a plain data protection question, and it bites at a much smaller scale. A solo developer with fifty users is fully exposed to it. It is also separate from the question of whether the government has to approve AI models before release, which it generally does not. Nobody approves your provider choice. You own it.

For US developers the analysis is looser but not absent. Client confidentiality, contractual obligations to your own customers, and any sector rules you operate under all survive the fact that the API was free.

What OpenRouter's free router changes

OpenRouter is a routing service that exposes many providers behind one API. In February 2026 it shipped openrouter/free, a single endpoint that selects from its current pool of zero-cost models. It removes the busywork of chasing whichever model is free this week.

It also concentrates the problem. You are no longer choosing one provider's policy. You are accepting whichever downstream provider the router picks.

OpenRouter's own documentation is candid about this. Account settings control whether you allow routing to providers that may train on your data, with separate switches for free and paid models. Flip training off and the free pool contracts, because many free endpoints exist precisely because the provider gets the data. The documentation also notes the setting "has no bearing on OpenRouter's own policies."

The honest reading is that a free model and a private model are usually the same slot, and you can have one of them.

So can you ship on a free tier?

Yes, for a specific and fairly wide set of projects. The test is not how serious the project is. It is whose text passes through the key.

Free tier is fine when:

1. You are prototyping with synthetic or your own data.
2. The input is already public, such as summarising a published article.
3. You are evaluating model quality before committing budget.
4. You are building an internal tool and every user knows the terms.

Move to a paid key when:

1. Any end user's text reaches the API.
2. You handle personal data of EU or UK residents.
3. You have promised customers confidentiality.
4. You are storing prompts that contain business-sensitive information.

The upgrade is cheaper than most people assume, because the paid tier bills on usage rather than a subscription. For a low-traffic app, buying out of the data clause can cost a few dollars a month. That is the actual price of the free tier, and it is worth paying early.

One practical wrinkle: Google no longer publishes free-tier rate limits in its documentation. The Gemini API rate limits page now directs developers to check their own quota in AI Studio. Free capacity is a moving target, which is another reason not to build a product's foundation on it.

The bottom line

Free AI API tiers are a genuinely good deal for prototypes and a bad deal for products. The rate limit is the advertised constraint. The data clause is the real one, and it is the only part of the plan that can create a legal problem rather than an engineering one.

Read the terms for the tier you are on, not the provider in general. On Google's API those are two different documents with opposite conclusions. Then decide whether the text flowing through your key is yours to trade.