If you’ve ever wanted to prototype an agentic workflow, try out a new model, or just play around with an LLM API without putting a card on file, you’ve probably hit the same wall: which providers actually offer real free access, and what are the real limits?
free-llm-api-resources by cheahjs answers exactly that. It’s a community-maintained reference — with 23.8K stars — that catalogs services offering free API access to LLMs, split into two groups: providers with a genuine free tier, and providers that give you trial credits to get started.
What makes it useful isn’t the idea (we all know free tiers exist), but that the list stays updated and includes the actual rate limits per provider, so you know upfront what you’re working with.
What’s in the list
The repo covers the providers you’d expect and several you might not:
- Google AI Studio — Gemini models with generous free quotas. The detail worth knowing: outside UK/CH/EEA/EU, your data is used for training on the free tier.
- Groq — fast inference, free tier with per-model limits (roughly 30 requests/minute, 14,400 requests/day on documented models).
- Cerebras — free access with token caps per minute and per day; requires phone number verification.
- OpenRouter — routes to multiple providers and exposes a set of free models.
- Cloudflare Workers AI, HuggingFace Serverless Inference (limited to models under 10GB, with some exceptions), Mistral, GitHub Models, and more.
It also lists trial credit providers (Together, Fireworks, Baseten, among others) for when you’ve exhausted the free tiers but still aren’t ready to commit.
A couple of things the maintainer makes explicit, and worth repeating: don’t abuse these services — overuse is exactly how free tiers disappear. And the list deliberately excludes anything sketchy, like services reverse-engineering existing chatbots.
Why it’s worth bookmarking
Two reasons. First, free tiers and their limits change constantly — providers raise caps, discontinue models, add verification requirements. A static blog post on this topic becomes outdated the week after it’s published. This repo regenerates from source data and gets updated by the community, so it’s the closest thing to a source of truth.
Second, rate limits per provider are the part you really need when planning. Knowing one provider gives you 1M tokens/day versus another’s 60K tokens/minute is the difference between a workflow that runs and one that throttles halfway through.
The honest caveat: treat the numbers from any article (including this one) as a snapshot in time. By the time you read this, some limit may have shifted. Go to the repo to see the current state — that’s exactly why it exists.
How to use it
There’s nothing to install. Open the README, find a provider whose limits fit your use case, sign up, grab an API key. Most speak OpenAI-compatible API format, so switching providers in your code is usually just a matter of changing the base URL and key.
For prototyping, testing a model before paying for it, or running side projects that don’t justify a bill, it’s hard to beat free.
Which of these providers are you already using for your side projects? Any that should be on the list?