Skip to content
Free LLM Today

Compare free AI APIs before you sign up

9 providers of AI language models (LLMs) side by side: how much each one gives away, whether it asks for a credit card, and what it does with your prompts.

  • Free to use, no sign-up
  • No API key asked
  • Checked

9 of 9 providers

Showing 9 of 9 providers. The large number is what the provider’s own page promised on the date under it. “Not published” means the page gives no figure; we do not borrow one from elsewhere. See every limit in one table

Every limit and rule in one table

The same providers, line by line. Each value has a check mark with the day it was read: tap it to open the official page.

9 of 9 providers · scroll the table sideways for all columns

Free LLM API tiers: limits and terms per provider, each with its source and check date
ProviderFree allowanceRequests / minRequests / dayCard to startTrains on your dataCommercial useOpenAI SDKImage input
GroqFree tier
30 req/min · 1,000 req/day

Chat models; 8K tokens/min and 200K tokens/day.

30
1,000
Not published
Not published
Not verified
Yes
Yes
Google Gemini APIFree tier
Not published

Limits are per project and shown only inside Google AI Studio.

Not published
Not published
Not required
Yes
Restricted
Yes
Yes
OpenRouterFree tier
20 req/min · 50 req/day

1,000 req/day once the account has bought at least 10 credits.

20
50
Not required
Your setting
Not verified
Yes
Yes
Cloudflare Workers AIFree tier
10,000 Neurons/day

Resets at 00:00 UTC. Going above it needs Workers Paid ($0.011 per 1,000 Neurons).

Not published
Not published
Not verified
No
Not verified
Yes
Not verified
Hugging Face Inference ProvidersFree tier
$0.10 of credits/month

Listed as “subject to change”. Past it, usage needs purchased credits.

Not published
Not published
Not verified
Not verified
Not verified
Yes
Not verified
NVIDIA NIMFree tier
Not published

The product page publishes no rate limit and no credit amount.

Not published
Not published
Not verified
Not verified
Not for production
Not verified
Not verified
CerebrasTrial only
$5 trial credit, 30 days

Granted “after adding a verified payment method”.

5
Not published
Required
Not verified
Not verified
Yes
Not verified
GitHub ModelsRetired
Not published

Service retired on 30 July 2026.

Not published
Not published
Not published
Not published
Not published
Not published
Not published
Mistral La PlateformeNot confirmed
Not verified
Not verified
Not verified
Not verified
Not verified
Not verified
Not verified
Not verified

74 of 108 cells above were read on the provider’s own page in the last 30 days (oldest check: ). Each check mark opens that page. The other 34 say “not verified” instead of repeating a number from someone else’s article.

Not sure which one? Answer three questions

The answer uses only what each provider’s own page confirms today.

Will you send images?
Must your prompts stay out of training?
Requests per day

Confirmed on the official pages for what you asked:

  • Groq
    30 req/min · 1,000 req/day

    Chat models; 8K tokens/min and 200K tokens/day.

  • Google Gemini API
    Not published

    Limits are per project and shown only inside Google AI Studio.

  • OpenRouter
    20 req/min · 50 req/day

    1,000 req/day once the account has bought at least 10 credits.

  • Cloudflare Workers AI
    10,000 Neurons/day

    Resets at 00:00 UTC. Going above it needs Workers Paid ($0.011 per 1,000 Neurons).

  • Hugging Face Inference Providers
    $0.10 of credits/month

    Listed as “subject to change”. Past it, usage needs purchased credits.

  • NVIDIA NIM
    Not published

    The product page publishes no rate limit and no credit amount.

A provider is left out only when its own page contradicts what you asked. Speed is not a question here because we do not measure it.

One request, any provider

Providers with a compatible endpoint take the same code. Switch the provider: only the base URL, the model id and the name of the key variable change. Base URLs side by side

Language
base_url
https://api.groq.com/openai/v1
model
openai/gpt-oss-120b
key from
$GROQ_API_KEY
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.groq.com/openai/v1",
    api_key=os.environ["GROQ_API_KEY"],
)

reply = client.chat.completions.create(
    model="openai/gpt-oss-120b",
    messages=[{"role": "user", "content": "Reply with the word: ok"}],
)
print(reply.choices[0].message.content)
Endpoint read on · console.groq.com

The key is read from an environment variable on your machine. This site has no field for keys and never asks for one.

Go straight to your question

One page per question, each with the providers that confirm the answer.

What changed lately

Free tiers get cut, turned into trials or retired. Each entry links to the page that confirms it.

  • · ReducedCerebras lists a time-limited trial, not a recurring free tier
  • · PolicyGemini API free-tier limits are not published per model
  • · BaselineFirst collection of OpenRouter zero-price models
  • · RetiredGitHub Models fully retired
Full change log, with sources

Why you can trust these numbers

Three steps stand behind every value on this site.

  1. We open the provider’s own page

    Rate-limit, pricing and terms pages of each provider. No blog posts, no forum threads.

  2. We write the number and the day

    Every value keeps the address it was read on and the date, and shows both next to it.

  3. After 30 days it expires

    A value nobody re-read turns into “not verified” by itself, in your browser.

A date and a source next to every number
Each value stores the official page it was read on and the day. The check mark next to it opens that page.
Old checks expire by themselves
A value read more than 30 days ago is shown as “not verified”. The page computes this from today’s date in your browser, without waiting for an update of the site.
“Not published” is an answer
When the provider’s page does not state a limit, the card and the table say so. We do not fill it with a figure from a blog post.
Terms next to limits
Card requirement, use of your prompts for training and commercial-use restrictions sit beside the rate limits.
No key ever asked
There is no field for an API key on this site. The code sample reads it from an environment variable on your machine.

How to read a free tier

Four things that change what a “free” number is worth.

The unit changes from provider to provider

Groq and OpenRouter count requests. Groq also counts tokens per minute and per day, and for chat that is the limit you hit first. Cloudflare counts Neurons, its own unit of compute. Hugging Face gives a monthly credit in dollars. Two free tiers cannot be ranked by one number; compare each against the traffic you expect.

The limit belongs to an account or a project

Google applies Gemini API limits per project, not per key. OpenRouter counts all free model variants of an account together. Creating more keys does not multiply a free allowance.

The terms cost more than the limit

A free tier is paid for in some other way: with your prompts, as on the Gemini API free tier, where content is used to improve Google products; or with a restriction, as on NVIDIA NIM, where free access stops at production. Read the data and commercial-use columns before the rate limit.

Free tiers end

GitHub Models was retired on 30 July 2026. Cerebras now describes a 30-day trial with a payment method where older articles describe a free tier. Keep the provider behind configuration, with a second one ready, and follow the change log.

Questions and answers

What people ask before they pick a free tier.

Which LLM APIs have a free tier right now?

On 3 Oct 2026 we confirmed a free offer on the official pages of Groq, Google Gemini API, OpenRouter, Cloudflare Workers AI, Hugging Face Inference Providers, NVIDIA NIM. Cerebras lists a 30-day trial with a payment method instead of a recurring free tier, and GitHub Models was retired on 30 July 2026.

How do I know the limits on this page are current?

Every cell has a check mark with the day it was read and a link to the provider’s page. After 30 days without a new check, the cell stops showing the value as current and says “not verified”. Your browser makes that calculation from today’s date.

Why do some cells say “not published”?

Because the provider’s own page does not state the value. Google, for example, does not list free-tier limits per model: its rate-limits page says they can be viewed in Google AI Studio. We print that instead of a figure taken from a third-party article.

Do I have to paste my API key here?

No. There is no field for a key anywhere on this site. The code sample reads the key from an environment variable on your computer.

Is a free tier enough for production?

Rarely. Free limits are low, can change without notice, and some terms exclude production: NVIDIA limits free NIM access to development, testing, research and evaluation, and Google requires paid services for apps offered to users in the EEA, Switzerland or the UK. Check the commercial-use column before building on one.

How often is the OpenRouter free model list updated?

It is read from the OpenRouter models API by a scheduled job once a day, and each page shows the collection time in UTC. The copy on this page was collected on 3 Oct 2026 and lists 22 zero-price models, 17 of them with an id ending in “:free”.