Uncensored Endpoint Pricing: Per-Token Math Explained

The two rates, and the one formula

Token type	Price per 1M tokens	Price per 1,000 tokens
Input (your prompt, including history)	$0.15	$0.00015
Output (the model's reply)	$0.50	$0.0005

Every request costs input tokens / 1,000,000 x $0.15 + output tokens / 1,000,000 x $0.50. The response includes a usage object with the exact prompt and completion token counts, so you never have to guess after the fact. This page works the numbers by hand so that you can predict spend before you write code.

Output tokens cost a little over three times as much as input tokens. That shapes how you design prompts: a long, detailed prompt with a short answer is cheap, and a short prompt that asks for a very long answer is where the money goes. Both extremes are still small amounts per request, as the examples show.

Two things are worth fixing in your head before the examples. First, tokens are not words. English prose runs at roughly three quarters of a word per token, so 1,000 tokens is about 750 words, and code or text in other languages usually needs more tokens for the same length. Second, billing happens per request on exact counts from the server, not on estimates from your side. Your own estimate only needs to be close enough to choose a top-up amount and set an alert, and the examples below give you the shape of that estimate for four common workloads: a chat turn, a long session, a single huge prompt and a batch job.

Example 1: one chat turn

Take a typical bot or chat app exchange. The system prompt plus recent history is 2,000 tokens, and the model answers with 500 tokens.

• Input: 2,000 / 1,000,000 x $0.15 = $0.0003
• Output: 500 / 1,000,000 x $0.50 = $0.00025
• Total: $0.0003 + $0.00025 = $0.00055

A thousand such turns cost $0.55. The $0.50 trial credit pays for about 909 of them (0.50 / 0.00055 = 909.09, rounded down), which is a generous test. A $20 top-up covers about 36,000 turns (20 / 0.00055 = 36,363, rounded down). If your app does a few hundred chat turns a day, the monthly bill stays in the single dollars.

Example 2: a roleplay session where history grows

The surprise for most newcomers is that the model has no memory. Chat clients resend the whole conversation on every turn, so the input side grows with each message. Suppose a roleplay session starts at 500 prompt tokens, and every turn adds 450 tokens to the history (a 150-token user message plus a 300-token reply). The model writes 300 tokens per turn, for 50 turns, with no trimming.

Turn n sends 500 + 450 x (n - 1) input tokens. Summed over 50 turns:

• Input tokens: 50 x 500 + 450 x (0 + 1 + ... + 49) = 25,000 + 450 x 1,225 = 576,250
• Output tokens: 50 x 300 = 15,000
• Input cost: 0.57625 x $0.15 = $0.0864375
• Output cost: 0.015 x $0.50 = $0.0075
• Total: about $0.094

Under ten cents for fifty turns, and the story is far longer than the model's replies suggest. Notice, though, that input is 576,250 tokens against only 15,000 output tokens. Input grows with the square of the number of turns, so a session of 200 turns does not cost four times as much as 50 but closer to sixteen times as much on the input side. Trimming old turns, or replacing them with a short summary, flattens that curve. The 100,000-token window is a ceiling, not a target.

Example 3: one huge prompt

Now the opposite case: you paste a long document and ask for a summary. The prompt fills the window with 100,000 tokens, and the reply is 2,000 tokens. Remember that the window counts prompt and completion together, so 100,000 input tokens plus 2,000 output tokens would be slightly over the limit. Use a prompt of 98,000 tokens for the real thing:

• Input: 98,000 / 1,000,000 x $0.15 = $0.0147
• Output: 2,000 / 1,000,000 x $0.50 = $0.001
• Total: $0.0157

About a cent and a half for reading roughly 70,000 words. A hundred of those cost $1.57. Long inputs are the cheapest kind of work on this price list, because the input rate is the lower one. If you can express a job as lots of reading and a short answer, such as classification, extraction or summarising, it fits the pricing well.

Example 4: a batch job

Say you need to tag 10,000 support tickets. Each ticket plus instructions is 1,500 input tokens, and the model returns a 200-token JSON object (you ask for JSON mode so the output is easy to parse).

• Input: 10,000 x 1,500 = 15,000,000 tokens = 15 x $0.15 = $2.25
• Output: 10,000 x 200 = 2,000,000 tokens = 2 x $0.50 = $1.00
• Total: $3.25, or $0.000325 per ticket

Cost is not the limit for a job like this; speed is. A full key allows 8 requests at once and 300 per minute, so 10,000 requests need at least 10,000 / 300 = 33.3 minutes at the request cap, whatever the concurrency. Run eight workers, retry any 429 after a short pause, and expect around 35 minutes to an hour depending on how long each answer takes. A trial key at 60 requests per minute would need nearly three hours for the same job, and it only holds $0.50, so run batches on a funded full key.

Top-ups, bonuses and what you get

Credit is prepaid. You add any whole amount from $20 to $500, and a bonus applies at three thresholds: 5% from $50, 10% from $100 and 15% from $200. Payment is in crypto, either USDT on TRC20 or USDC on Base.

You pay	Bonus	Credit received	Chat turns at $0.00055
$20	none	$20.00	about 36,300
$50	5%	$52.50	about 95,400
$100	10%	$110.00	about 200,000
$200	15%	$230.00	about 418,100
$500	15%	$575.00	about 1,045,400

The last column divides credit by the Example 1 turn cost and rounds down; your own turn cost will differ. The bonus tiers reward bigger top-ups, but there is no reason to over-fund: credit never expires, yet it is only worth buying what you expect to use. A sensible path is the trial first, then $20 to check real usage, then a larger top-up once you know your monthly rate. See the pricing page for the current tiers.

What you do not pay for, and how to watch spend

There is no subscription, no seat fee and no minimum monthly spend. Failed or refused requests are not charged, so a 429 or an error response costs nothing. Prompts are not used for training, which is a privacy point rather than a cost one, but worth knowing when you compare options.

To keep control of spending, use what the API already gives you:

• Read X-Balance-USD from the response headers and log it with each request, or alert when it falls below a level you pick.
• Set max_tokens for jobs where output length matters. It is optional, but it puts an upper bound on the more expensive side of the bill.
• Trim chat history to a fixed number of messages, as in Example 2.
• Count usage.prompt_tokens and usage.completion_tokens per user or per feature, and multiply by the two rates to get a cost per feature.

Other providers publish their own prices and change them from time to time, so check their current pricing yourself if you are comparing. For a quick start, claim the one-click trial from the API key page, run your real prompts, read the usage counts, and put them into the formula at the top of this page. The documentation shows the request format, and the production checklist covers key handling once you go live.

More guides

• Uncensored Endpoint: Base URL, Model ID and Limits
• Uncensored Endpoint Telegram Bot: Build It in Python
• Uncensored AI API Key: Production Checklist