EN ▾

GPT Model API Pricing: Per-Token Costs, Worked Examples

Pricing for a GPT-style chat API comes down to tokens in and tokens out. Here are our rates, the math for real workloads, and the habits that keep a bill small.

Updated

Isometric machine dispensing glowing token cubes, illustrating per-token API pricing

Key points

  • Rates: $0.15 per 1M input tokens and $0.50 per 1M output tokens.
  • Pay as you go from prepaid credit; no subscription, and credit never expires.
  • 100,000 requests of 1,500 tokens in and 400 out cost $42.50.
  • Failed or refused requests are not charged.

How the bill is built

If you are comparing gpt model api pricing, start with the unit: the token. A token is a chunk of text, often a short word or part of a longer one. A rough rule for English is that 1,000 tokens equal about 750 words, though code and other languages use more tokens per word. The API response tells you the exact counts in its usage field, so you never have to guess after the fact.

Every request is billed on two counts. Input tokens are everything you send: system prompt, conversation history and the new message. Output tokens are everything the model writes back. Output costs more per token than input. Our rates are $0.15 per 1M input tokens and $0.50 per 1M output tokens.

Worked examples

Let us compute a few common shapes. Each follows the same formula: tokens divided by 1,000,000, times the rate.

A short question. 200 tokens in, 150 out. Input: 200 / 1,000,000 x $0.15 = $0.00003. Output: 150 / 1,000,000 x $0.50 = $0.000075. Total: $0.000105. About 9,500 such requests cost one dollar.

A chat turn with history. 3,000 tokens in, 400 out. Input is $0.00045, output is $0.0002, total $0.00065. A thousand of these cost $0.65.

A document task. 20,000 tokens in (a long text to summarize), 1,000 out. Input is $0.003, output is $0.0005, total $0.0035. Summarizing 1,000 documents of that size costs $3.50.

A full context window. The window is 100,000 tokens, prompt plus completion. A request with 90,000 in and 10,000 out costs $0.0135 + $0.005 = $0.0185, under two cents.

ShapePer requestPer 1,000 requests
200 in, 150 out$0.000105$0.105
3,000 in, 400 out$0.00065$0.65
20,000 in, 1,000 out$0.0035$3.50
90,000 in, 10,000 out$0.0185$18.50

Monthly estimates

Projects think in months, so scale up. Suppose a support tool sends 100,000 requests per month, averaging 1,500 input and 400 output tokens.

  • Input: 100,000 x 1,500 = 150,000,000 tokens, which is 150M x $0.15 = $22.50
  • Output: 100,000 x 400 = 40,000,000 tokens, which is 40M x $0.50 = $20.00
  • Total: $42.50 per month

A smaller app with 5,000 requests a month at the same shape costs $2.13 (a twentieth of $42.50 is $2.125). A creative-writing tool where users generate long passages may flip the ratio: 20,000 requests with 1,000 in and 2,000 out give 20M input tokens at $3.00 plus 40M output tokens at $20.00, so $23.00.

Notice that output dominates when replies are long, and input dominates when history is long. Know which one your app is before you try to optimize.

Growth is linear, which makes planning easy. If traffic doubles, the bill doubles: 200,000 requests of the same shape cost $85.00. If you cut average input from 1,500 to 1,000 tokens by trimming history, the input line falls from $22.50 to $15.00 per 100,000 requests, a saving of $7.50 a month without touching quality of the replies.

Credit, top-ups and bonuses

You pay from prepaid credit, so spending is capped by what you loaded. The credit never expires, and there is no subscription fee.

Top-ups are any whole amount from $20 to $500, by crypto (USDT on TRC20, USDC on Base). Bigger top-ups carry a bonus: +5% from $50, +10% from $100, +15% from $200. That means $50 becomes $52.50 of credit, $100 becomes $110, and $200 becomes $230. Using the $42.50 monthly example, a $50 top-up covers a little over a month of that workload, with $10 left over once the bonus is counted.

New users can start with a $0.50 trial credit, claimed with one click after sign-up. It is valid for 7 days and one per person. See the pricing page for the current table.

One practical note on sizing: the minimum top-up is $20, which at the 3,000-in, 400-out chat shape ($0.00065 per request) is about 30,769 requests. For most prototypes that is weeks of testing, so there is little reason to start higher.

Seven ways to keep the bill down

The rates are low, but waste still adds up at volume. These are the changes that save the most, roughly in order of effect.

  1. Trim history. Keep the last N turns and a short summary of earlier ones. Input tokens grow with every turn otherwise.
  2. Shorten the system prompt. It is resent on every request. A 1,500-token prompt sent 100,000 times is 150M input tokens, or $22.50.
  3. Set max_tokens where it fits. It is optional here, but a cap prevents runaway replies.
  4. Use stop sequences. They end generation once you have what you need.
  5. Ask for compact output. JSON mode with short keys beats prose when a program reads the result.
  6. Cache your own answers. If many users ask the same thing, store the reply instead of paying again.
  7. Watch the balance. The X-Balance-USD response header shows what is left on every call.

Budget checklist before launch

Before an app goes in front of users, run through this list.

  • Measure average input and output tokens from at least 50 real requests.
  • Compute the worst case: a full 100,000-token window costs about $0.0185 for a 90,000-in, 10,000-out request. Decide whether your app allows that.
  • Cap user input length so one pasted document cannot multiply your costs.
  • Set max_tokens on endpoints where long replies are not needed.
  • Handle HTTP 429 with a retry that waits, using the Retry-After value when present. A full key allows 8 concurrent requests and 300 per minute.
  • Alert on balance. Read X-Balance-USD and notify yourself when it passes a threshold, so you can top up before users notice.

Comparing quotes from other providers

We will not quote other vendors' prices here, since they change and we cannot vouch for them. You can still compare fairly using the same method, and it takes ten minutes.

First, collect each provider's input rate and output rate per 1M tokens from their current pricing page. Second, take one real prompt and one real reply from your app and count their tokens. Third, run the formula for each provider with your counts. Fourth, multiply by your monthly request count.

Watch for differences that do not show up in the rate line. Some providers count tokens differently, so the same text can be a different number of tokens on each service. Some charge for features such as extra context, and some have minimum purchases or subscriptions on top of usage. Ours has none of these: no subscription, a $20 minimum top-up, credit that never expires.

Last, compare what the model does, not only what it costs. A cheaper call that refuses your request is worth nothing, and a pricier one that answers may be cheaper per useful result. For adult fiction and roleplay, test your own prompts on each service before you decide.

Measure your real cost in code

Estimates are only as good as your token counts, so log real ones. This snippet prices each response from its own usage data.

from openai import OpenAI

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

RATE_IN = 0.15   # per 1M input tokens
RATE_OUT = 0.50  # per 1M output tokens

resp = client.chat.completions.create(
    model="uncensored",
    messages=[{"role": "user", "content": "Summarize why logging usage matters."}],
    max_tokens=200,
)
u = resp.usage
cost = u.prompt_tokens / 1_000_000 * RATE_IN + u.completion_tokens / 1_000_000 * RATE_OUT
print(u.prompt_tokens, u.completion_tokens, round(cost, 6))

Run it over a day of real traffic, average the results, and multiply by your expected volume. That number beats any chart, including the ones above.

What the price does not include

Price is one input to a decision. This is a text-only API: no image, audio or video generation, no embeddings and no fine-tuning. It runs one open-weight model from an independent provider, not a model from OpenAI or another vendor, and it is not a router across vendors. If you need those features, you need another service. Check their current pricing before you compare.

What you get at this price is an OpenAI-compatible endpoint, streaming, tool calling, JSON mode, a 100,000-token context window, and a model tuned to answer without content refusals for lawful adult use. Prompts are not used for training. The hard limit is no sexual content involving minors.

To start, grab a key on the API key page and follow the docs to swap your base URL.

Questions and answers

How much does it cost per 1M tokens?

$0.15 per 1M input tokens and $0.50 per 1M output tokens. You pay only for what you use, from prepaid credit.

Is there a monthly fee?

No. There is no subscription. Credit never expires, so an unused month costs nothing.

Are errors and refused requests billed?

No. Failed or refused requests are not charged.

What is the smallest top-up?

$20. You can top up any whole amount from $20 to $500, with +5% from $50, +10% from $100 and +15% from $200.

Can I test before paying?

Yes. New users can claim a $0.50 trial credit with one click on the API key page. It is valid for 7 days, one per person.

Your key is one form away

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

Get API key