Is there a completely free LLM API?

Yes, with limits. Google's Gemini API has a free tier for its Flash and some Pro models, Cloudflare Workers AI gives 10,000 Neurons a day at no charge, OpenRouter offers free model variants with daily caps, and local models through Ollama cost nothing but your own hardware. None of them is unlimited, and several use your prompts to improve their products.

Does the Gemini API free tier use my data?

Yes. Google's pricing page states that content sent on the free tier is used to improve its products, while paid-tier content is not. Don't send confidential data on the free tier.

Is the OpenAI API free?

Not for ongoing use. OpenAI's quickstart lets you run a free test request, and accounts in supported countries start on a Free usage tier, but meaningful usage requires prepaid credits.

Is the Claude API free?

No. Anthropic gives new Console accounts a small amount of credit for testing, and after that usage is billed per token from prepaid credits.

Are the free API keys posted on GitHub safe to use?

No. They're almost always leaked or stolen keys that belong to someone else. Using them is unauthorized access, the keys get revoked quickly, and anything you send through them goes to an account you don't control.

Free LLM APIs in 2026: Verified Free Tiers, Real Limits and the Catches

Short answer: the main free LLM APIs as of October 2026 are:

Google's Gemini API free tier: Flash and some Pro models, with rate limits shown in AI Studio, and your data is used to improve Google's products.
Cloudflare Workers AI: 10,000 Neurons per day.
OpenRouter's :free model variants: daily request caps.
Cerebras: a one-off $5 trial credit.
Local models through Ollama: free, running on your own hardware.

OpenAI and Anthropic only offer a small starting allowance. Everything below was checked against each provider's own documentation on October 15, 2026. Free tiers change often, so the date matters.

At a glance

Provider	What's free	Limits	Catch
Google Gemini API	Free tier on Gemini 3.8 Flash, 3.6 Flash, 3.5 / 3.1 Flash-Lite, 3 Flash Preview, 2.5 Pro, 2.5 Flash and others	Per-model RPM, TPM and RPD, shown in AI Studio. Daily limits reset at midnight Pacific	Free-tier content is used to improve Google's products. Gemini 4 Argon, 3.1 Pro Preview and image models aren't on the free tier
Cloudflare Workers AI	10,000 Neurons per day, on the Workers Free and Paid plans	Neurons measure GPU compute and convert to tokens per model	Mostly open-weight models. Paying for more requires Workers Paid
OpenRouter	Model variants whose ID ends in `:free`	Per-minute and per-day request caps. The daily cap rises once your all-time credit purchases pass a threshold	Free variants come and go, and the upstream provider may log prompts
Cerebras	$5 trial credit after adding a verified payment method	Free Trial: e.g. 5 RPM, 30K TPM and 1M tokens/day for `gpt-oss-120b`	Credits expire after 30 days. There's no permanent free tier
Ollama (local)	Any open model you can run	Your hardware	Needs a capable GPU or Apple silicon for good speed
OpenAI	A free test request; Free usage tier in supported countries	Low limits until you buy credits	Real usage needs prepaid credits
Anthropic	Small credit for new Console accounts	Evaluation-tier limits	Real usage needs prepaid credits

Google Gemini API: the most generous

Create a key in Google AI Studio and you can call most Flash-class Gemini models at no cost. According to Google's pricing page (October 2026):

Free: Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 3 Flash Preview, 2.5 Pro, 2.5 Flash, plus several TTS and Live models.
Not free: Gemini 4 Argon, Gemini 3.1 Pro Preview and the image-generation models.
Data: content sent on the free tier is used to improve Google's products. On the paid tier it isn't.
Limits are set per model and per project, not per key, and Google doesn't publish fixed numbers. Your actual RPM, TPM and RPD appear on the Rate limit page in AI Studio. Preview and experimental models get tighter limits.

Gemini's free tier also speaks the OpenAI format (https://generativelanguage.googleapis.com/v1beta/openai/), so most OpenAI-compatible tools can use it. See OpenAI-compatible APIs explained.

If you mainly want free coding help in the terminal, Gemini CLI with a personal Google sign-in allows up to 1,000 model requests a day. See Install Gemini CLI.

Cloudflare Workers AI

Every Cloudflare account gets 10,000 Neurons per day free, on both the Free and Paid Workers plans. Neurons are Cloudflare's unit of GPU compute. Each model's page converts them to tokens. A small model like @cf/meta/llama-3.2-1b-instruct costs 2,457 Neurons per million input tokens, so the free allowance goes a long way on small models and much less far on big ones. Beyond 10,000 Neurons a day you need Workers Paid, at $0.011 per 1,000 Neurons.

OpenRouter free variants

OpenRouter lists some models in a free version whose ID ends in :free. They share per-minute and per-day request caps, and the daily cap is higher once your all-time credit purchases pass a threshold set by OpenRouter. You can check your current counter with GET https://openrouter.ai/api/v1/key (the free_model_daily_requests field). Free variants appear and disappear, and each is served by an upstream provider with its own data policy. Read the policy on the model page before sending anything private.

Cerebras: a trial, not a tier

New Cerebras accounts get $5 in credits after adding a verified payment method. The credits expire 30 days after they're granted. During the trial, limits are tight, for example 5 requests per minute and 1M tokens per day on gpt-oss-120b. Cerebras states plainly that there's no permanent free tier.

Local models: free if you have the hardware

Running open models yourself with Ollama costs nothing per token, and Ollama exposes an OpenAI-compatible API at http://localhost:11434/v1/. The trade-offs are your hardware's speed, memory limits on model size, and that open models still trail the best hosted ones on hard coding and reasoning tasks.

OpenAI and Anthropic: starter allowances only

OpenAI: the quickstart lets you run a free test request, and accounts in supported countries start on a Free usage tier. For real work, add prepaid credits. Tiers rise from $5 in purchases. See How to get an OpenAI API key.
Anthropic: new Claude Console accounts get a small free credit for testing. After that, it's prepaid. See How to get a Claude API key.

Their cheapest current models are close to free anyway: GPT-6 Luna and Claude Haiku 5.5 both cost $0.10 per million input tokens, so a few dollars of credit covers a lot of experimentation.

Free tools with built-in models

Some apps bundle free model access, which is handy if you just want to use AI rather than build with an API:

Gemini CLI: free Gemini access with a Google sign-in.
OpenCode (terminal coding agent): ships a rotating set of free models. Some "contributor" models use your prompts for training.
Cherry Studio (desktop chat app): a built-in CherryAI provider with free models. See the Cherry Studio guide.

Avoid "free API key" lists

Search for "free OpenAI API key" or "free Claude API key" and you'll find GitHub repos and posts full of keys. They're leaked or stolen keys belonging to other people. Using them is unauthorized access to someone else's account, they get revoked within hours, and you'd be sending your prompts through an account you don't control. If you need cheap access, use one of the real free tiers above, or a low-cost model.

How to choose

You want…	Use
The most free capacity for a hosted frontier-class model	Gemini API free tier (keep sensitive data off it)
Free AI inside your own serverless app	Cloudflare Workers AI
To try many open models through one API	OpenRouter `:free` variants
Privacy and no per-token cost	Ollama on your own machine
Production reliability on a small budget	A paid low-cost model (GPT-6 Luna, Claude Haiku 5.5, Gemini Flash-Lite), or an API gateway

When a free tier stops being enough, an API gateway is one way to keep a single key across providers. api.foo, which publishes these docs, is one, but it's paid. Compare the per-token price with going direct before you choose.