EN ▾
Get API key
Uncensored LLM API pricing guide

API Pricing per Million Tokens: Worked Cost Examples

People who search "openai api pricing per million tokens 2026" usually want a number they can drop into a budget. This page gives that number for our hosted API, $0.15 per 1M input tokens and $0.50 per 1M output tokens, and prices real workloads token by token.

Updated

Balance scale weighing stacks of token cubes, illustrating per-million-token pricing
Key points
  • Our rates are $0.15 per 1M input tokens and $0.50 per 1M output tokens, paid from prepaid credit.
  • A typical chat request of 2,000 tokens in and 500 out costs $0.00055, so $1 covers about 1,800 of them.
  • There is no free unlimited tier. New users get a $0.50 trial, valid for 7 days, with no card.
  • To compare any other provider, run the same formula with their current per-million rates.

What you pay per million tokens here

This site is not OpenAI and does not publish OpenAI's prices. The domain name describes the question people ask, and the answer on this page is about the API we run: a hosted, OpenAI-compatible text API with one open-weight model, uncensored, operated by an independent provider. If you need OpenAI's own rates, open their pricing page and read the current numbers there, because they change and a copy here would go stale.

Our pricing has exactly two numbers.

Token typePrice per 1M tokensPrice per 1,000 tokens
Input (your prompt, history, system message)$0.15$0.00015
Output (what the model writes back)$0.50$0.0005

That is the whole price list. There is no subscription, no seat fee and no separate charge for tool calling, JSON mode or streaming. You prepay credit, requests draw it down, and the credit never expires. Requests that fail or are refused are not charged. The same numbers appear on the pricing page, which is the place to check if anything here ever looks out of date.

The formula, and three jobs priced with it

Every bill comes from one line:

cost = (input tokens x 0.15 + output tokens x 0.50) / 1,000,000

Here are three jobs of different shapes, each computed from that line.

A single chat request

Say a user message plus the earlier conversation adds up to 2,000 input tokens, and the model answers with 500 tokens. Input: 2,000 x 0.15 / 1,000,000 = $0.0003. Output: 500 x 0.50 / 1,000,000 = $0.00025. Total: $0.00055. One dollar buys 1 / 0.00055, or about 1,818 requests of that size.

Summarizing a long document

You send an 80,000-token document and ask for a 1,000-token summary. Input: 80,000 x 0.15 / 1,000,000 = $0.012. Output: 1,000 x 0.50 / 1,000,000 = $0.0005. Total: $0.0125, a little over one cent. The input side dominates here, which is typical for reading-heavy work.

A batch of 10,000 requests

Each request has 1,500 input tokens and 600 output tokens. The batch totals 15,000,000 input tokens and 6,000,000 output tokens. Input: 15 x $0.15 = $2.25. Output: 6 x $0.50 = $3.00. Total: $5.25 for the batch, or $0.000525 per request. Notice that the output, only 29 percent of the tokens, is 57 percent of the bill. Output tokens cost 0.50 / 0.15, about 3.3 times as much as input tokens, so long answers move the total faster than long prompts.

A 20-turn chat, priced turn by turn

Chat apps are where people most often underestimate cost, because every turn resends the whole history. Here is a clean example. A user and the model exchange 20 turns. Each turn adds 150 tokens from the user and 350 tokens from the model, 500 tokens in total.

On turn k the input is everything said so far plus the new user message: 500 x (k - 1) + 150 tokens. Add that up over 20 turns:

  • History part: 500 x (0 + 1 + 2 + ... + 19) = 500 x 190 = 95,000 tokens.
  • New user messages: 150 x 20 = 3,000 tokens.
  • Total input: 98,000 tokens.
  • Total output: 350 x 20 = 7,000 tokens.

Cost: 98,000 x 0.15 / 1,000,000 = $0.0147 for input, plus 7,000 x 0.50 / 1,000,000 = $0.0035 for output. The whole session costs $0.0182, under two cents. The final turn sends 9,650 input tokens, far inside the 100,000-token context window.

The lesson is that input grows with the square of the number of turns while output grows linearly. At 20 turns the input already outweighs the output by 14 to 1 in tokens. At 100 turns of the same size it would be 500 x 4,950 + 15,000 = 2,490,000 input tokens against 35,000 output tokens, which is $0.3735 plus $0.0175, or $0.391 for the session. Still small, but it shows where the money goes. Trimming or summarizing old turns is the one habit that matters at scale.

What a top-up buys

Credit is added in whole dollar amounts from $20 to $500. Bigger top-ups carry a bonus: +5% from $50, +10% from $100 and +15% from $200. Using the 2,000-in, 500-out request from above at $0.00055 each:

You payBonusCreditRequests of that size
$20none$20.00about 36,300
$50+5%$52.50about 95,400
$100+10%$110.00200,000
$200+15%$230.00about 418,000
$500+15%$575.00about 1,045,000

You can pay by crypto (USDT on TRC20, USDC on Base). Because the credit does not expire, there is no penalty for buying a larger block than you need this month. If you only want to test, the trial described below costs nothing. The API key page is where both the trial claim and the top-up live.

There is no free unlimited tier

Search results for pricing queries are full of the phrase "free tier", so it is worth being blunt. We do not offer an unlimited free plan. What exists is a one-time trial: after you sign up (with Google, or with email and password), you can claim $0.50 of credit with one click on the API key page. It is limited to one per person, valid for 7 days and needs no card.

Fifty cents is enough to learn a lot. At $0.00055 per request it covers about 909 requests of the size above, or roughly 27 full 20-turn sessions at $0.0182 each. Trial keys have lower limits than full keys, namely 2 concurrent requests and 60 requests per minute, against 8 concurrent and 300 per minute for a full key. That is plenty to wire up a client and test your prompts, and not enough for production traffic.

When the trial ends, you either top up or stop. Nothing is billed automatically, because there is no card on file for a trial and no subscription behind it.

Comparing against another provider's per-million rates

Per-million pricing looks easy to compare and often is not. If you are weighing us against a different API, collect the following from that provider's own current pricing page instead of relying on a blog post, including this one:

  • The input price and the output price, as two separate numbers.
  • Whether the price you are reading belongs to the exact model you would use.
  • Whether cached or batched requests are priced differently, and whether you can actually use that path.
  • Whether failed or refused requests are billed.
  • Whether prepaid credit expires and whether there is a minimum top-up.
  • The rate limits for a new account, since a low limit can cost you more in engineering time than a higher token price.

Then price your own workload with the formula, not an average. Take 100 real requests from your logs, sum the input and output tokens, and multiply. A support bot with short answers and a story generator with long answers can differ by a factor of three on the same input price. We keep a side-by-side on the ChatGPT API cost page, written with the same rule: we state our numbers and point you to theirs.

Keep the product difference in view too. Our model is tuned to answer without content refusals for lawful adult use, it is text only and it has a 100,000-token context window. If you need image input, embeddings or a specific vendor's flagship model, a lower price per million tokens will not make this API the right choice.

Keeping the bill predictable

Three controls keep spend where you expect it. First, set max_tokens on endpoints where you know the answer length, since output is the expensive side. Second, cap how much history you resend. Third, log the usage numbers that come back with every response so you can reconcile them against your balance.

The response carries a usage object, and the headers include X-Balance-USD, so you can watch credit fall in real time. This script prices one call exactly:

from openai import OpenAI

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

raw = client.chat.completions.with_raw_response.create(
    model="uncensored",
    messages=[{"role": "user", "content": "Summarize the plot of Hamlet in five sentences."}],
    max_tokens=300,
)
resp = raw.parse()
u = resp.usage
cost = (u.prompt_tokens * 0.15 + u.completion_tokens * 0.50) / 1_000_000
print(u.prompt_tokens, "in,", u.completion_tokens, "out, cost", round(cost, 6))
print("balance after call:", raw.headers.get("X-Balance-USD"))

The same check works from the shell if you want to see the headers directly:

curl -si https://api.openaiapipricing.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"uncensored","messages":[{"role":"user","content":"Say hi"}],"max_tokens":20}' \
  | grep -i -E "x-balance-usd|x-request-id|ratelimit"

The quickstart covers the rest of the setup, including streaming and tool calling.

Questions and answers

What is the price per million tokens on this API?

$0.15 per 1M input tokens and $0.50 per 1M output tokens. Both are billed from prepaid credit that never expires.

Do you publish OpenAI's prices?

No. This is an independent hosted API and not an OpenAI product. For OpenAI's rates, read their current pricing page.

Is there a free tier?

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

Am I charged for failed or refused requests?

No. Failed or refused requests are not charged.

How do I estimate cost before I commit?

Take 100 real requests, add up input and output tokens, and apply input x 0.15 plus output x 0.50, divided by 1,000,000.

Your key is one form away

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