EN ▾
Get API key

Free LLM API to Use: Trial, Limits and Real Costs

There is no free unlimited tier here, but you can test the API at no cost with a $0.50 trial and no card. This page shows what that trial buys in real requests and what happens after it runs out.

Updated

Isometric illustration of a trial credit coin flowing through a pipeline of tokens into a terminal
01

Key points

  • New users can claim $0.50 of trial credit with one click on the API key page, valid for 7 days, no card needed.
  • Prices are $0.15 per 1M input tokens and $0.50 per 1M output tokens, billed from prepaid credit.
  • The trial covers roughly 900 medium chat requests, which is enough to judge the model on your own prompts.
  • Top-ups start at $20, credit never expires, and failed or refused requests are not charged.

The short answer

If you are looking for a free LLM API to use, here is the honest version. This service does not offer an unlimited free tier. What it offers is a trial: after you sign up, one click on the API key page adds $0.50 of credit to your account. It needs no card, it is limited to one claim per person, and it stays valid for 7 days.

That sounds small, and it is, but tokens are cheap. At $0.15 per million input tokens and $0.50 per million output tokens, fifty cents goes a long way when you are testing a prompt, wiring up a client or checking whether the model answers the way you need. The sections below put real numbers on it so you can decide before you sign up.

The model behind the endpoint is a single open-weight model, tuned to answer without content refusals for lawful adult use. It is text only, with a 100,000 token context window, and the model ID is uncensored. Anything beyond that, such as images or embeddings, is not part of the product.

What the trial gives you

The trial is meant for evaluation, not production traffic. You get the same API as paying users, with a few tighter limits.

  • Credit: $0.50, claimed once per person.
  • Validity: 7 days from the claim.
  • Rate limits: 2 concurrent requests and 60 requests per minute. A full key allows 8 concurrent requests and 300 per minute.
  • Sign-up: Google, or email and password.
  • Features: streaming, tool calling and JSON mode all work on the trial key.

The 2-request concurrency cap matters if you test with a script that fires requests in parallel. A streamed response holds its slot until the stream ends, so two long streams in flight will make a third request fail with HTTP 429 and the error code concurrency. Run your tests one or two at a time and you will not hit it.

When the 7 days pass, any unused trial credit expires. If you only wanted a quick look, that is fine. If you plan to keep going, add a top-up first and the API keeps working with the same key.

What a request costs, with the math

Billing is per token and split into two meters: tokens you send (input) and tokens the model writes back (output). The formula is simple.

cost = input_tokens x 0.15 / 1,000,000 + output_tokens x 0.50 / 1,000,000

Here are four request sizes worked through exactly, and how many of each the $0.50 trial credit pays for.

Request typeInput / output tokensCost per requestRequests per $0.50
Short question500 / 200$0.0001752,857
Typical chat turn2,000 / 500$0.00055909
Document summary20,000 / 1,000$0.0035142
Full context window100,000 / 4,000$0.01729

Check one row by hand. A typical chat turn sends 2,000 input tokens: 2,000 x 0.15 / 1,000,000 = $0.0003. It receives 500 output tokens: 500 x 0.50 / 1,000,000 = $0.00025. Together that is $0.00055, and $0.50 divided by $0.00055 is about 909 requests.

The last row is the extreme case. Filling the whole 100,000 token window costs 100,000 x 0.15 / 1,000,000 = $0.015 for the prompt, plus $0.002 for 4,000 output tokens, which is $0.017 in total. Even a prompt that large is under two cents.

Failed and refused requests are not charged, so retrying after an error does not eat your credit. Every response also carries an X-Balance-USD header, which lets you watch the remaining balance from your own code instead of guessing.

Making the trial last

Output tokens cost more than three times what input tokens cost, so the cheapest habit is to ask for shorter answers. A few other changes help as well.

Cap the output while you test

The max_tokens parameter is optional, because the completion can use whatever is left of the 100,000 token window. During testing, set it to something like 300. A model that rambles for 2,000 tokens costs four times as much as one stopped at 500, and you rarely need the long version to check that your integration works.

Trim the history you resend

Chat APIs are stateless. Each request carries the whole conversation, so input cost grows with every turn. A chat that starts at 500 tokens and gains 600 per exchange has sent 500 + 1,100 + 1,700 + 2,300 + 2,900 = 8,500 input tokens after five turns, not 2,900. Dropping old turns or summarising them keeps that curve flat.

Test with fixed settings

The seed and temperature parameters are passed through. When you compare two prompt versions, hold them constant so you are not paying for noise. Use a low temperature for extraction and classification work, and raise it only for creative output.

Use a script, not a UI

Send a batch of your own real prompts from a short script and read the answers side by side. Twenty well chosen prompts tell you more than two hundred idle chats, and at the medium size above they cost about $0.011 in total (20 x $0.00055).

When prepaid credit makes sense

Once the trial confirms the model suits your use case, the next step is a top-up. There is no subscription and no monthly minimum. You add credit, you spend it per token, and it never expires, so unused balance simply waits.

Top-ups run from $20 to $500 in any whole amount. Larger ones carry a bonus: +5% from $50, +10% from $100 and +15% from $200. That works out as follows.

  • $20 gives $20.00 of credit.
  • $50 gives $52.50.
  • $100 gives $110.00.
  • $200 gives $230.00.
  • $500 gives $575.00.

To judge what $20 means for your project, take a chat product where each request averages 3,000 input and 600 output tokens. One thousand requests use 3M input tokens ($0.45) and 0.6M output tokens ($0.30), so $0.75 per thousand. At that rate $20 covers about 26,666 requests. Even a busy side project can run for weeks on a first top-up.

You can pay with crypto (USDT on TRC20, USDC on Base). The pricing page lists the current terms, and the overview of the pay-per-token API explains the product in more detail.

Other ways to get a free LLM API

The trial is not the only route, and depending on your goal it may not be the best one. Being straight about that is more useful than pretending otherwise.

Run an open-weight model yourself. If you have a capable GPU, local servers such as Ollama or llama.cpp let you run models at no per-token cost. You pay in hardware, electricity and setup time, and larger models need more memory than most laptops have. For a prototype on small models this is a reasonable choice. For anything that needs steady throughput or a bigger model, a hosted endpoint is simpler.

Use other providers' free tiers. Several vendors offer limited free usage for their own models. Those terms change often, and they usually come with the vendor's own content rules, so check their current pricing and policies before you build on one. If your project needs a model that does not refuse lawful adult requests such as fiction or roleplay, a free tier from a mainstream vendor may not fit regardless of price.

Start here with the trial, then decide. Because the endpoint is OpenAI-compatible, switching later costs you one base URL and one key. Nothing in your code is locked in. The quickstart shows the full setup, and it takes only a few lines with any OpenAI SDK.

One boundary applies to every use: sexual content involving minors is not allowed, and no setting changes that.

02

Questions and answers

Is there a free LLM API to use with no limits?

No. There is a one-time $0.50 trial credit valid for 7 days, with no card required. After that you pay per token from prepaid credit.

How do I claim the free trial credit?

Sign up with Google or with email and password, open the API key page, and click once to claim $0.50. It is limited to one claim per person.

What happens when the trial credit runs out?

Requests stop until you add credit. The same key keeps working after a top-up of $20 or more, and nothing in your code needs to change.

Do failed requests use up my credit?

No. Failed or refused requests are not charged, so you can retry after an error without losing balance.

Are my prompts used for training?

No. Prompts are not used for training.

Your key is one form away

Create an account, copy the key, change the base URL. That is the whole setup.

Get API key