EN ▾
Uncensored LLM API starting at $0.15 per 1M tokenshttps://api.cheapllmapis.com/v1

Cheap LLM API: Real Costs and How to Cut Them

Cheap only means something once you add up a real workload. This page does that for a per-token API at $0.15 input and $0.50 output per 1M tokens, then shows how to shrink the total.

Glass jar of glowing tokens beside a falling line, illustrating lower per-token costs

Key points

  • Prices are $0.15 per 1M input tokens and $0.50 per 1M output tokens, prepaid, with no subscription.
  • A 50,000-request month at 1,500 input and 250 output tokens costs $17.50.
  • Trimming history and capping output cuts that same month to $9.75.
  • Failed requests are not charged, and credit never expires.

What makes an API cheap

A cheap LLM API is one where your total monthly bill stays low for the work you actually do, and a low headline rate alone does not give you that. Three things decide that: the per-token prices, the extra charges around them (subscriptions, minimums, expiring credit), and how many tokens your application sends. This API covers the first two simply. It charges $0.15 per 1M input tokens and $0.50 per 1M output tokens, with no subscription, no expiry on credit, and no charge for failed or refused requests.

The third factor is yours to control, and it is where most savings come from. The next sections work through a concrete workload and then the changes that reduce it. The model ID is uncensored, a single open-weight model for lawful adult use, text only.

A month of real traffic, calculated

Take an app with 50,000 requests a month. Each request sends about 1,500 input tokens (system prompt, history, new message) and gets back about 250 tokens.

  • Input: 50,000 x 1,500 = 75,000,000 tokens = 75 x $0.15 = $11.25
  • Output: 50,000 x 250 = 12,500,000 tokens = 12.5 x $0.50 = $6.25
  • Total: $17.50 for the month

That fits inside the minimum $20 top-up with $2.50 left over, and unused credit stays on the account. Notice that input is 64% of the bill here, even though it is the cheaper rate, because there is so much more of it. Output tokens cost more per token ($0.50 is about 3.3 times $0.15), but input usually dominates in chat apps because history is resent every turn.

For comparison, a lighter workload of 5,000 requests with the same sizes costs $1.75. A heavier one, 500,000 requests, costs $175. The cost scales in a straight line, so you can multiply to estimate your own case.

Five ways to lower the bill

1. Trim the history

Do not resend a chat from the beginning forever. Keep the system prompt and the last handful of messages, and summarise anything older in a sentence or two. The context window allows 100,000 tokens, which is room for a lot, but room is not a reason to fill it.

2. Cap the output

Set max_tokens when a reply has a known ceiling, and ask for brevity in the system prompt. It is optional, and without it the completion can use the rest of the window. A cap of 150 instead of an average of 250 tokens saves $0.00005 per request, which becomes $2.50 over 50,000 requests.

3. Use stop sequences and JSON mode

For machine-read output, response_format with {"type":"json_object"} keeps the reply to the object you asked for instead of chatty framing. A stop sequence ends generation as soon as the part you need is written.

4. Fix retries

When you hit HTTP 429, wait before retrying. The Retry-After: 5 header tells you how long on temporary overload. Failed requests are not charged, but duplicate successful ones are, so do not fire the same call twice from two code paths.

5. Batch within your concurrency

A full key allows 8 concurrent requests and 300 per minute. Running a queue at 8 workers uses what you have paid for without tripping concurrency errors.

The same month after tuning

Apply the first two changes to the 50,000-request month. History trimming brings input down from 1,500 to 800 tokens per request, and a 150-token output cap brings the average reply down from 250 to 150.

  • Input: 50,000 x 800 = 40,000,000 tokens = 40 x $0.15 = $6.00
  • Output: 50,000 x 150 = 7,500,000 tokens = 7.5 x $0.50 = $3.75
  • Total: $9.75, down from $17.50, a 44% reduction

Whether shorter replies are acceptable depends on your product, so test the quality change before you roll it out. Here is a minimal history trimmer with a cap in place:

from openai import OpenAI

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

MAX_TURNS = 6  # keep the last 6 messages besides the system prompt

def trim(history):
    system, rest = history[0], history[1:]
    return [system] + rest[-MAX_TURNS:]

history = [{"role": "system", "content": "You are a concise assistant."}]
history.append({"role": "user", "content": "Plan a three-day trip outline."})

resp = client.chat.completions.create(
    model="uncensored",
    messages=trim(history),
    max_tokens=300,
)
history.append({"role": "assistant", "content": resp.choices[0].message.content})

The response also reports usage, so log it per request and you will see the effect of every change in real numbers.

Long documents and big contexts

The 100,000-token window makes whole-document jobs possible, so price one before you run a batch. A call with 99,000 input tokens and 1,000 output tokens costs 99,000 x $0.15 / 1M = $0.01485 plus 1,000 x $0.50 / 1M = $0.0005, so $0.01535. Running that on 1,000 documents costs $15.35. If the document is only 30,000 tokens and the summary 500, the call costs $0.0045 + $0.00025 = $0.00475, and 1,000 documents cost $4.75.

Split-and-summarise pipelines often cost more than one large call, because every chunk repeats the instructions and each pass adds its own output tokens. Compare both on a sample before choosing.

How long does a top-up last

Use the tuned month, $9.75, as a yardstick. A $20 top-up covers a bit over two months of it (20 / 9.75 = 2.05). A $100 top-up adds 10% and becomes $110, which covers about 11 months (110 / 9.75 = 11.3). Credit never expires, so a larger top-up costs nothing extra in waiting; the bonus is the reason to consider one, and you can always start at $20.

Funding the account

You prepay. Any whole amount from $20 to $500 works, and bigger top-ups add credit: +5% from $50, +10% from $100, +15% from $200. So $50 becomes $52.50, $100 becomes $110 and $200 becomes $230. Pay by by crypto (USDT on TRC20, USDC on Base). The full table is on the pricing page.

Before paying anything, you can use the free trial. After sign-up, one click on the API key page claims $0.50 of credit, one per person, valid for 7 days, with no card. At the earlier sizes of 1,500 input and 250 output tokens, a request costs $0.000225 + $0.000125 = $0.00035, so $0.50 covers about 1,400 of them. Trial keys are limited to 2 concurrent requests and 60 per minute.

What cheap does not include

Be clear about the scope before you build. The API offers text chat completions and a models list. There is no image or audio generation, no embeddings and no fine-tuning, so a project that needs those pays another provider for them. The model is not the official API of any big vendor and not a router. And sexual content involving minors is never permitted.

If you are weighing this against a Claude-style flagship, the article on a cheap Claude API alternative covers the trade-offs. Other vendors change prices often, so check their current pricing instead of trusting a table copied from somewhere. When you are ready to wire it up, the quickstart guide takes the base URL, the key and the model name, and the rest is a normal OpenAI client call.

Questions and answers

How much does one request cost?

It depends on tokens. A request with 1,500 input and 250 output tokens costs $0.00035, calculated at $0.15 and $0.50 per 1M tokens.

Is there a free tier?

There is a $0.50 trial credit, claimed with one click on the key page after sign-up. It is valid for 7 days and needs no card.

Do unused credits expire?

No. Credit never expires, so you can top up $20 and spend it at your own pace.

Are retries billed?

Failed or refused requests are not charged. A request that succeeds is billed even if your code also sent a duplicate.

Your key is one form away

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

Get API keyRead the docs