Key points
- Monthly cost = requests x tokens per request x rate, split into input and output.
- Example loads here run from about $0.21 to $130 a month on this API.
- For OpenAI's own rates, check their current pricing page and plug the numbers into the same formula.
- Output tokens cost more than input here, so reply length is the main lever.
The formula
Whatever provider you use, a monthly bill comes from one calculation. Count the tokens you send (input) and the tokens you receive (output), then multiply each by its rate per million.
Monthly cost = (input tokens / 1,000,000 x input rate) + (output tokens / 1,000,000 x output rate)
Rates differ between providers and change over time, so we do not quote OpenAI or anyone else here. Open their pricing page, copy the input and output rates, and use them in place of ours below. The tokens column will not change. On this service the rates are $0.15 input and $0.50 output per 1M tokens, with no subscription and no minimum monthly spend.
How to estimate tokens without guessing
A token is roughly three quarters of an English word, so 1,000 words is about 1,300 tokens. For a real number, send a few typical prompts and read the usage field in each response. This script does that and projects a month from your own sample:
from openai import OpenAI
client = OpenAI(base_url="https://api.uncensoredgptapi.com/v1", api_key="YOUR_KEY")
samples = [
"Explain what a webhook is in two paragraphs.",
"Write a polite reply declining a meeting.",
"Summarize: the quarterly report shows rising costs and flat sales.",
]
tin = tout = 0
for s in samples:
r = client.chat.completions.create(
model="uncensored",
messages=[{"role": "user", "content": s}],
)
tin += r.usage.prompt_tokens
tout += r.usage.completion_tokens
per_request = tin / 1e6 * 0.15 / len(samples) + tout / 1e6 * 0.50 / len(samples)
requests_per_month = 5000
print(f"per request ${per_request:.6f}, month ${per_request * requests_per_month:.2f}")
Remember that chat history counts as input every turn. A conversation resends earlier messages, so your average input per request is larger than any single user message.
Four example workloads
| Workload | Requests / month | Input / output tokens each | Total on this API |
|---|---|---|---|
| Personal assistant | 600 | 1,200 / 350 | $0.21 |
| Content batch (articles) | 2,000 | 800 / 1,500 | $1.74 |
| Support bot | 9,000 | 2,500 / 300 | $4.73 |
| Agent or coding tool | 100,000 | 6,000 / 800 | $130.00 |
The math, line by line
- Personal assistant. Input 600 x 1,200 = 720,000 tokens = 0.72 x $0.15 = $0.108. Output 600 x 350 = 210,000 = 0.21 x $0.50 = $0.105. Total $0.213.
- Content batch. Input 2,000 x 800 = 1.6M = $0.24. Output 2,000 x 1,500 = 3M = $1.50. Total $1.74.
- Support bot. Input 9,000 x 2,500 = 22.5M = $3.375. Output 9,000 x 300 = 2.7M = $1.35. Total $4.725.
- Agent tool. Input 100,000 x 6,000 = 600M = $90. Output 100,000 x 800 = 80M = $40. Total $130.
What the numbers tell you
Three patterns show up. First, the light cases are tiny: a personal assistant costs about twenty cents. Second, the content batch is output-heavy, so output at $0.50 per 1M is 86% of its bill. Third, the agent tool is input-heavy, because every step re-reads context, and that is where cost grows fast.
If you want to reduce the monthly figure, the order of impact is usually: cut repeated context (trim history, summarize old turns), shorten replies where long ones add nothing, then batch low-priority work. Changing the model matters less than changing the token count, since the tokens column drives the bill.
From users to a monthly bill
Most people start from users, not requests. Suppose a chat feature has 500 daily active users, each sending 6 messages a day for 30 days. That is 500 x 6 x 30 = 90,000 requests. Say each request averages 1,800 input tokens (history included) and 250 output tokens.
Input is 90,000 x 1,800 = 162M tokens, or 162 x $0.15 = $24.30. Output is 90,000 x 250 = 22.5M tokens, or 22.5 x $0.50 = $11.25. The month comes to $35.55, which is about $0.07 per active user.
Check the rate limit too. 90,000 requests a month is about 3,000 a day, or roughly two a minute on average. Even a peak ten times that is about 21 a minute, far below the 300 per minute a full key allows. The tighter constraint is concurrency: 8 requests at once. Streaming replies hold a slot until they finish, so a long answer ties up capacity longer than a short one.
Guardrails so the bill stays predictable
Open-ended usage is the real risk, not the rate. A loop that retries without a limit, or a user who pastes a 90,000-token document ten times, can spend more than your plan assumed. A few cheap controls fix most of it:
- Read
X-Balance-USDfrom the response headers and alert when it drops under a set amount. - Cap input length per message and cap history length per chat.
- Set
max_tokensfor features with a known answer size. - Log
usageper feature so you can see which one drives cost. - Back off on 429 responses instead of retrying immediately.
With prepaid credit, the worst case is bounded by your balance. That is a different risk profile from a post-paid invoice, where a bug can produce a surprise charge after the fact.
Pricing your own app on this API
Credit is prepaid. You top up any whole amount from $20 to $500, and bonuses add +5% from $50, +10% from $100 and +15% from $200. A $100 top-up becomes $110 of credit, which covers the support bot above for roughly 23 months. Credit never expires, so a month with no traffic costs nothing, and failed or refused requests are not charged.
Payment is by crypto (USDT on TRC20, USDC on Base). The pricing page lists the options. For a side-by-side look at how GPT-style access is priced, see GPT model API pricing.
Test the numbers before you commit
New users can claim $0.50 of trial credit on the API key page after sign-up. It needs no card and lasts 7 days. At the support-bot rate of about $0.0005 per request, the trial covers around 1,000 requests, which is plenty to validate your token estimates.
Trial keys allow 2 concurrent requests and 60 per minute; full keys allow 8 concurrent and 300 per minute. If you only need to swap a base URL in existing code, our GPT API base URL guide has the steps, and the docs cover every endpoint.
One boundary: this is a text API serving a single open-weight model tuned to answer without content refusals for lawful adult use. It is not the OpenAI API and has no image, audio or embedding endpoints, so if your monthly estimate includes those, price them separately.

