What it does

This is an awesome-list that catalogs LLM providers offering permanent or trial free tiers for text inference. Each entry links directly to the provider’s API key page and lists base URLs, model names, context windows, output limits, modalities, and rate limits in a uniform table format. The README splits providers into two camps: model owners (Google, Mistral, DeepSeek, etc.) and third-party inference hosts (Groq, Cerebras, Cloudflare, GitHub Models, etc.).

The interesting bit

The list treats “free” as a spectrum and documents the strings attached—trial credits that expire, geographic exclusions, non-commercial clauses, even whether your prompts get used for training. That transparency is rarer than it should be in this space. Most endpoints are OpenAI SDK-compatible, so swapping keys is often just a config change.

Key highlights

• 15+ provider APIs (AI21, Cohere, xAI, Zhipu, etc.) with no-credit-card signups
• 10+ inference providers including Cerebras (~2,600 tok/s), Groq (LPU-backed), and GitHub Models (45+ models for all users)
• Rate limits, context caps, and output maxima explicitly tabulated per model
• Footnotes flag gotchas: EU/UK blocks, training-data opt-outs, deprecation dates
• All endpoints OpenAI SDK-compatible unless otherwise noted

Caveats

• “Permanent free” is provider-defined; some are time-capped trials (DeepSeek: 30 days, AI21: 3 months)
• Rate limits can be severe—Zhipu allows one concurrent request, Gemini free tier is 5 RPM
• README is truncated in the source; full Groq table and any entries beyond it are cut off

Verdict

Handy if you’re prototyping, building a router, or just refuse to give OpenAI your billing info. Less useful if you need production SLAs or predictable costs—free tiers are the weather, not climate.