Free AI API keys: what is actually free in 2026

Most BYOK apps are free or cheap to buy, but you still need a key that works. Here is what you can use without paying in October 2026, what the limits are, and the catch with each.

The short list

Option	What is free	Main limit	The catch
Google Gemini free tier	Gemini Flash, Flash-Lite and 2.5 Pro models	Requests per minute and per day, per project	Free-tier content is used to improve Google's products
OpenRouter :free models	A rotating set of open models	20 per minute; 50 per day (1,000 after buying $10 of credits)	Smaller or older models, availability varies
Groq free plan	Fast open models (for example gpt-oss and Qwen)	Per-model daily and per-minute caps	Open models only; limits shown in your Groq console
Local models (Ollama, LM Studio)	Everything, offline	Your computer's memory and speed	Smaller models than the cloud frontier

Google Gemini: the best free cloud option

Create a key in Google AI Studio (step-by-step in our Gemini API key guide). On 2026-10-23 Google offered a free tier for Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 2.5 Pro, 2.5 Flash and 2.5 Flash-Lite. Limits are counted per Google Cloud project and reset daily at midnight Pacific time; AI Studio shows your exact numbers.

The catch is data: Google says content sent on the free tier is used to improve its products for most models. Fine for experiments and public text, not for private documents.

OpenRouter free models

Any model on OpenRouter whose ID ends in :free costs nothing. You can make 20 requests per minute and 50 per day; buy $10 of credits once and the daily limit rises to 1,000, while the free models stay free. One OpenRouter key works in every app that supports OpenRouter, which is a lot of them.

The catch: the free selection is mostly open models and changes over time, and free providers can be slower or busier.

Groq free plan

Groq runs open models such as OpenAI's gpt-oss and Qwen on its own hardware and is known for very fast answers. It has a free plan; your per-model limits (requests and tokens per minute and per day) are shown on your account's limits page. Good for voice and real-time apps that support Groq.

Local models: free and private

Ollama and LM Studio run open models on your own computer. Apps connect to a local server (Ollama listens on http://localhost:11434, with an OpenAI-compatible endpoint at /v1), so there is no key, no per-token cost and nothing leaves your machine. You need enough memory for the model you pick. See local LLM apps.

If you already pay for something

• Claude Max includes monthly API credits ($100 on Max 5x, $200 on Max 20x) usable on the Claude Platform. See the Claude API key guide.
• Some BYOK apps also offer a few free model calls of their own before you add a key; check each listing.

No longer free

• GitHub Models was fully retired on July 30, 2026, including its free playground and API.
• Mistral's free plan now covers its chat app (Vibe), not API access, according to its pricing page.

When paying a little is better

Free tiers are fine for trying apps. For daily use, a few dollars on a cheap model is often the better deal: with DeepSeek Flash or OpenAI's gpt-5.6-luna, a typical chat turn costs about a tenth of a cent, so $5 lasts thousands of messages, without free-tier limits or data terms.