Which free LLM APIs work without a credit card — and what are the published limits?

[xyzs996]

Short answer: 15 of them, and none of the 15 asks for a card. They are the "provider free tier" rows in this repo — permanent tiers, not trial credits that expire. Every limit below is copied from the provider's own documentation page, and the whole set was re-read on 2026-07-25.

The ones with a published number

Provider	Published free limit	Card	Source
SiliconFlow	1,000 RPM (free models are fixed; limits are per account, not per key, and each model is counted separately)	Not required	docs
GroqCloud	30 RPM / 1,000 requests per day on llama-3.3-70b-versatile; llama-3.1-8b-instant gets the same 30 RPM but 14,400 per day	Not required	docs
Cohere	20 requests/minute on chat, and a trial key is additionally capped at 1,000 API calls per month	Not required	docs
SambaNova Cloud	20 RPM / 20 requests per day / 200,000 tokens per day — the free tier applies exactly while no payment method is linked	Not required	docs
Fireworks AI	10 RPM across the whole account when there is no payment method and no credits	Not required	docs
Cloudflare Workers AI	10,000 Neurons/day, resetting 00:00 UTC — a compute unit, not a request count, so how many calls you get depends on the model and prompt	Not required	docs
Hugging Face Inference Providers	$0.10 in monthly routing credits for signed-in free users ($2.00 for PRO)	Not required	docs

The ones that are free but publish no number

Google Gemini API, Z.AI (GLM-4.7-Flash and GLM-4.5-Flash are listed at $0 for input, cached input and output), Mistral, Alibaba Cloud Model Studio, Moonshot (Kimi), Novita, Pollinations.AI, Ollama Cloud.

These are genuinely free to call. They just don't put a per-minute or per-day figure on a public page — Gemini, for instance, publishes the tier structure and then sends you to AI Studio for the numbers actually active on your project. So this repo records "dynamic / model-dependent" rather than inventing a figure.

What this list is not

• Not a key giveaway. Nothing here distributes anyone's API key. Every row links to the provider's own signup page and you make your own.
• Not "free forever" guaranteed. GitHub Models is in the dataset as a retiring tier — it shut to new users on 2026-07-30. That row is kept precisely so you can see that this happens.
• Not the same as cheap enough to run. Trial credits (Cerebras, Vercel AI Gateway, IBM watsonx.ai) and metered services (Together, Nebius, Perplexity, DeepInfra, Chutes) are in the dataset but deliberately kept out of the 15 — they are a different thing and get their own table.

Full filterable directory, all 26 entries: https://xyzs996.github.io/free-llm-api/ · how each claim was checked: https://xyzs996.github.io/free-llm-api/methodology.html

One thing that would help: if a limit above is already stale on your account — you hit a wall the table doesn't predict, or you got more than it says — reply with the provider and what you actually saw. A limit that moved is the single hardest thing to catch from the outside, and one report is enough to send me back to the source page.