Free LLM APIs · checked 2026-10-22
Cloudflare Workers AI: what “free” actually gets you
Workers AI runs open models on Cloudflare's network. Every account, free or paid, gets a daily free allowance measured in Neurons, Cloudflare's own unit of compute.
- Free allowance
- 10,000 Neurons/day
- Card needed
- No
- Trains on prompts
- No
- Main catch
- Neurons, not requests: how far they go depends on the model
- Base URL
https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/v1
Limits
- 10,000 Neurons per day on both the Free and the Paid plan, resetting at 00:00 UTC.
- On the Free plan, requests fail once the day's Neurons are used up. On the Paid plan, usage beyond them is billed.
The fine print
- A Neuron is not a request or a token. Each model converts tokens to Neurons at its own rate, so 10,000 Neurons buys many small-model replies or few large-model ones.
- The OpenAI-compatible URL includes your account id, and the key is a Cloudflare API token with Workers AI permission.
Your data
- Cloudflare does not use content processed by Workers AI to train AI models.
- For standard inference requests, inputs and outputs aren't kept once the response is returned. Features you turn on that store data (AI Gateway logging or caching, batch jobs) keep it under their own terms.
Get started
- Create a Cloudflare account and an API token with Workers AI permission.
- Use base URL https://api.cloudflare.com/client/v4/accounts/{ACCOUNT_ID}/ai/v1 with an OpenAI-compatible client.
Questions
- How much is free on Cloudflare Workers AI?
- 10,000 Neurons a day on every plan, resetting at 00:00 UTC. How many requests that is depends on the model.
- Does Cloudflare train on Workers AI prompts?
- No. Cloudflare says it doesn't use content processed by Workers AI to train AI models, and doesn't keep prompts for standard requests.
Sources
Read on 2026-10-22. Providers change limits and terms often; the linked pages are what applies.
Which free models are worth using today, and who serves them: state of free LLM APIs and the open ranking on GitHub. One endpoint across free models, with your own Groq or OpenRouter key if you have one: ZeroLimitAI.