One person’s field guide to free LLM APIs

What it does

Free-LLM-Collection is a hand-curated directory of no-cost LLM APIs that expose an OpenAI-compatible endpoint. The maintainer—a single person—ranks providers by whether they throttle requests rather than tokens, then by model freshness and usable rate limits. It covers roughly two dozen services, from major hubs like OpenRouter and HuggingFace to smaller gateways such as LLM7 and ChatAnywhere.

The interesting bit

The selection logic inverts typical free-tier anxiety: it prefers rate-limited access over token-limited access, on the theory that a slow pipe you can actually use beats a generous cap you burn through in one call. Even the summary table was generated by an LLM and rendered by a tool called Taple, so the catalog is organized by the same compute it lists.

Key highlights

• Concrete rate limits where known: 2 QPS for the lite endpoint, 1000 RPM for SiliconFlow and InceptionLab, and daily points budgets for ChatAnywhere.
• Quirks noted up front: Intern AI keys expire after six months, some SenseNova tiers share upstream limits, and 肖恩AI requires daily check-ins.
• Scope is strictly OpenAI-format APIs and text models that have gained “some popularity,” which keeps the list interoperable but narrow.
• Accepts community submissions, though the maintainer warns they can only update as time allows.

Caveats

• Several entries carry “Unknown” rate limits or warn of upstream shared limits, so the documented caps are incomplete.
• The scope is limited to OpenAI-format text models with “some popularity,” which excludes niche or multimodal offerings.
• The project is maintained by one person, so updates depend on their availability.

Verdict

Worth bookmarking if you want to prototype against real models without adding another line item to your cloud bill. Skip it if you need enterprise reliability or APIs that do not follow the OpenAI request shape.