for-the-zero/Free-LLM-Collection

What this project is

Free-LLM-Collection is a hand-maintained list of large language model (LLM) API providers that offer free tiers. It both collects free APIs and picks out which models are worth using from each provider.

Who runs it

A single person. The README states it is currently maintained by one individual who promises to keep updating it as best they can.

Selection criteria — the author ranks providers and models by these priorities:

1. Long-term viability first — services that are likely to stay free and available.
2. Rate limiting style — the author prefers providers that limit requests per minute over ones that cap total tokens.
3. Source diversity — more different providers is better.
4. Model freshness and quality — newer, better models are favored.
5. Sufficient rate limits — the free tier must be usable, not token-throttled to nothing.

A hard requirement — only APIs that follow the OpenAI API format are included. This makes swapping between providers easy for developers.

How the data is presented

The project includes a visual table image. The README notes this table's data was generated by an LLM and rendered via a tool called "Taple" (linked). Below the image is a full Markdown breakdown.

Provider entries — each provider listing includes:

• An API endpoint (the URL you'd call)
• Rate limits (how many requests per minute/day/hour, or points/tokens limits, or "unknown" where the author didn't know)
• Recommended models or available models (some entries separate these — "Rec." means the author's picks, "Ava." means all free options)
• Occasional tips (e.g., Intern AI keys last 6 months; Cohere lets you unlock a looser Production Key by attaching a payment method)

Providers covered — roughly 30 services, including well-known names like Google Gemini, HuggingFace, NVIDIA, Mistral, Groq, Cloudflare, Cohere, ModelScope, plus smaller or niche ones (onomeo, OrcaRouter, Agens AI, ChatAnywhere, etc.). Each has its own quota, from generous (SiliconFlow at 1000 requests per minute per model) to tight (OpenRouter at 20/day for free models).

Contributions — the author explicitly welcomes submissions of additional free APIs from the community.

Extras at the bottom

• Sources — the list is built from the author's own research plus user submissions.
• Star history chart — tracks GitHub stars over time.
• Leaderboard links — points readers to third-party benchmark sites (llm_benchmark, Artificial Analysis, LMArena) for comparing model quality.

What it is not — it is not a proxy or a unified API service. It is a directory: you still sign up with each provider yourself, get your own API key, and call each endpoint directly.