Free LLM API for Developers: Trial vs Pay as You Go

Short answer: a trial, then cents per call

Nothing here is free forever. What you get is a free start and a very small bill after it. Sign up with Google or with email and password, open the API key page, and claim $0.50 of trial credit with one click. No card is needed. The credit is valid for 7 days and each person gets one.

After that you pay as you go at $0.15 per 1M input tokens and $0.50 per 1M output tokens. There is no subscription and no monthly minimum. The smallest top-up is $20, and the credit does not expire, so you can spend it over a year if your project is slow.

What the $0.50 trial buys

Cost per request is simple: input tokens times $0.15, plus output tokens times $0.50, both divided by 1,000,000. Two examples show the range.

Check the first row: 1,500 x $0.15 / 1,000,000 = $0.000225, and 300 x $0.50 / 1,000,000 = $0.00015. The sum is $0.000375. For the second row, 20,000 input tokens cost $0.003 and 800 output tokens cost $0.0004, so $0.0034 in total.

In practice the trial is enough to wire up a client, test streaming, try tool calls and decide whether the model fits your project.

Why free tiers are hard to build on

Free tiers exist so you can try a product, and their limits are set with that purpose in mind. The details change often, so check each provider's current terms before relying on one. In general, developers hit the same few problems.

• Rate caps sized for a demo. A script that works at ten requests becomes unusable at a few hundred.
• Changing limits. A limit that is fine today can be tightened later, and your pipeline learns about it from an error.
• Content filters. Many hosted models refuse requests for fiction, roleplay or edgy test data. A refusal in the middle of a batch job is a bug you cannot patch.
• Unclear data handling. Some free offerings use prompts to improve their models. Read the terms.

None of this makes free tiers bad. For a weekend prototype they are fine. The trouble starts when the prototype becomes a service.

There is also a planning problem. A free tier gives you no way to forecast cost, because there is no price to multiply by. With per-token pricing you can write the budget on one line before you write any code, and that makes a project easier to defend to a client or a partner. If your usage grows tenfold, you know the bill grows about tenfold, and nothing else changes.

What to check in any API before you commit

Whether a service costs money or not, ask the same questions before you wire it into your code.

1. Is the protocol standard? This API is OpenAI-compatible, so switching away later means changing two lines, not rewriting a client.
2. What are the limits in numbers? Here they are published: 8 concurrent requests and 300 per minute on a full key, 2 and 60 on a trial key, 8 MB body, a 100,000-token window.
3. What happens when a call fails? On this service, failed or refused requests are not charged.
4. What happens to your prompts? Here they are not used for training.
5. Does the model refuse your use case? This one is tuned to answer without content refusals for lawful adult use, which matters for fiction, roleplay and test data. The only hard stop is sexual content involving minors.
6. Can you leave? Credit is prepaid and never expires, and there is no contract.

How to make the trial last

Small habits stretch $0.50 a long way.

1. Set max_tokens while testing. It is optional on this API, but a cap of 300 stops a runaway answer from costing ten times what you planned. Output is the expensive side at $0.50 per 1M.
2. Keep system prompts short. A 1,000-token system prompt sent with every call adds $0.00015 per request.
3. Use seed and low temperature for repeatable tests, so you do not re-run a call just to see a different answer.
4. Do not resend whole histories unless you need them. The window is 100,000 tokens, and every token in it is billed as input.
5. Watch X-Balance-USD in the response headers to see exactly what each test cost.

One more trick is to log resp.usage for every call during development. After a day of testing you will know your real average, and you can replace the guesses in the tables above with your own numbers. Many people find their prompts are two or three times longer than they assumed, mostly because of system prompts and pasted examples.

A budget for a side project

Numbers help when you decide whether a project is worth funding. Take a small app that makes 200 calls a day at the dev-loop size of 1,500 tokens in and 300 out. Each call costs $0.000375, so a day costs $0.075 and a 30-day month costs $2.25. A $20 top-up would pay for about nine months at that rate.

Now take a heavier tool that makes 50 calls a day with 20,000 tokens of pasted context and 800 tokens out. Each call costs $0.0034, so a day is $0.17 and a month is $5.10. Even a doubling of traffic keeps both projects in single-digit dollars per month.

The point is that token pricing at this level makes cost a minor line item for most side projects. What decides success is whether the model does the job and whether the limits fit your traffic.

A first call you can run now

The API is OpenAI-compatible, so the standard Python SDK works after changing the base URL and key.

Copyfrom openai import OpenAI

client = OpenAI(
    base_url="https://api.dolphinllm.com/v1",
    api_key="YOUR_KEY",
)

resp = client.chat.completions.create(
    model="uncensored",
    messages=[{"role": "user", "content": "Explain idempotency in two sentences."}],
    max_tokens=200,
)
print(resp.choices[0].message.content)
print(resp.usage)

The usage object gives you exact prompt and completion token counts. Multiply them by the rates above and you have the cost of that call. Streaming, JSON mode and tool calling are all available too; the docs have examples for each.

When to move from the trial to a top-up

Move on when the trial runs out or when you need more than 2 concurrent requests. A full key allows 8 concurrent requests and 300 per minute. Top-ups run from $20 to $500 in any whole amount. Larger ones come with a bonus: +5% from $50, +10% from $100 and +15% from $200. A $50 top-up gives $52.50 of credit, and $200 gives $230.

You can pay with crypto (USDT on TRC20, USDC on Base). Full figures are on the pricing page, and you can grab a key from the key page.

What this API is not

To save you a wasted evening: it is a text-only service. There is no image, audio or video generation, no embeddings and no fine-tuning. It is one open-weight model run by an independent provider, not the official API of any big vendor and not a router across several. If your project needs embeddings for search, you will need another service for that part. If it needs a chat model that follows instructions and does not refuse, this fits.

If you are new to the site, the page on using an LLM without login restrictions explains how sign-up and keys work in more detail.

More guides

• Open WebUI API Endpoints: Connect an Uncensored LLM
• How to Use an LLM Without Login Restrictions