AI Chatbot API Pricing: Cost per Conversation
AI chatbot API pricing is easiest to judge per conversation, not per token. At $0.15 per 1M input tokens and $0.50 per 1M output tokens, a 20-turn chat costs about $0.0068, and the math below shows why.
Updated

Key points
- A 10-turn chat costs about $0.0022, a 20-turn chat $0.0068 and an untrimmed 40-turn chat $0.0232.
- History resent on every turn makes input cost grow faster than the number of turns.
- Trimming history to the last 10 turns cuts a 40-turn chat from $0.0232 to $0.0127.
- Billing is prepaid, crypto only here, with no subscription and no expiry on credit.
Cost per conversation at a glance
Most ai chatbot api pricing pages quote a rate per million tokens, which tells you little about a real chat. Here is the answer in the unit you care about, using these assumptions: a 300-token system prompt, 40 tokens per user message and 120 tokens per reply.
| Conversation | Input tokens | Output tokens | Input cost | Output cost | Total |
|---|---|---|---|---|---|
| 10 turns | 10,600 | 1,200 | $0.00159 | $0.0006 | $0.00219 |
| 20 turns | 37,200 | 2,400 | $0.00558 | $0.0012 | $0.00678 |
| 40 turns, full history | 138,400 | 4,800 | $0.02076 | $0.0024 | $0.02316 |
| 40 turns, last 10 turns kept | 68,800 | 4,800 | $0.01032 | $0.0024 | $0.01272 |
Per thousand conversations, that is $2.19, $6.78, $23.16 and $12.72. The rates are $0.15 per 1M input tokens and $0.50 per 1M output tokens. Your numbers will differ with longer prompts or replies, so the next sections show where each figure comes from.
Why input dominates
A chat API is stateless. Each request must carry the system prompt and the whole visible history, because the server does not remember earlier calls. On turn k the input is the system prompt, the (k minus 1) earlier user-and-reply pairs and the new message.
With S as the system prompt, u as the user message, r as the reply and T as the number of turns, total input across the chat is T x S + T x u + (u + r) x T x (T minus 1) / 2. The last term grows with the square of T. For the 20-turn row: 20 x 300 = 6,000, plus 20 x 40 = 800, plus 160 x 190 = 30,400. The sum is 37,200 input tokens. Output is simply T x r = 2,400 tokens.
That square term is why doubling a conversation from 20 to 40 turns multiplies cost by about 3.4 (from $0.00678 to $0.02316), not by 2. It also explains why output price, though higher per token, rarely drives the bill. In the 20-turn chat, input is $0.00558 of $0.00678, or 82%.
A calculator you can reuse
Put your own numbers in this function. It counts the history exactly as a chat client would resend it, with an optional cap on kept turns.
def convo_cost(turns, sys_tok=300, user_tok=40, reply_tok=120, keep=None,
rate_in=0.15, rate_out=0.50):
inp = out = 0
for k in range(turns):
past = k if keep is None else min(k, keep)
inp += sys_tok + past * (user_tok + reply_tok) + user_tok
out += reply_tok
return inp, out, inp / 1e6 * rate_in + out / 1e6 * rate_out
print(convo_cost(20)) # (37200, 2400, 0.00678)
print(convo_cost(40)) # (138400, 4800, 0.02316)
print(convo_cost(40, keep=10)) # (68800, 4800, 0.01272)
Replace the constants with averages from real traffic. Each API response returns a usage object with prompt and completion tokens, and the X-Balance-USD header shows the balance, so you can verify the calculator against actual spend after a day of testing.
Five ways to lower the bill
- Cap history. Keep the last 10 turns. In the example it saves 45% on a 40-turn chat ($0.01272 against $0.02316).
- Summarise old turns. Replace turns older than the cap with a 100-word summary. You pay for one summary call and then send far fewer tokens.
- Shorten the system prompt. It is paid on every request. Cutting it from 300 to 200 tokens saves 100 x T input tokens per chat, which is 2,000 tokens or $0.0003 over 20 turns. Small per chat, real across a million chats.
- Set reply length rules. A line such as "Under 100 words" lowers output tokens and keeps the history from growing as fast.
- Do not retry blindly. Failed or refused requests are not charged, but a retry loop that resends big prompts after a timeout can still stack up slow, wasted work.
The page on chatbot on website cost adds hosting and development to the picture, since the API is only one line in the total.
From per-chat cost to a monthly budget
Multiply by volume. Ten thousand 20-turn conversations a month cost 10,000 x $0.00678 = $67.80. A hundred thousand 10-turn conversations cost 100,000 x $0.00219 = $219. Add a safety margin of 30% for longer messages than you guessed.
Top-ups are any whole amount from $20 to $500. A $20 top-up buys about 9,132 ten-turn chats or 2,949 twenty-turn chats at the sizes above. Bonuses stack on larger top-ups: +5% from $50, +10% from $100 and +15% from $200, so $200 buys $230 of credit. There is no subscription, and credit never expires, which means a quiet month costs nothing. Check the pricing page for the current form.
This site accepts crypto only: USDT on TRC20 or USDC on Base. Plan a small reserve in a wallet, because a top-up is not instant in the way a stored payment method would be, and you do not want a launch to stall on an empty balance.
Estimating tokens before you have traffic
You need token counts to use any of this, and on day zero you have no logs. Two rough rules get you started. English text averages around four characters per token, so a 160-character message is about 40 tokens. A page of 300 words is about 400 tokens. Other languages and code usually take more tokens per character, so raise the estimate by a quarter if your users write in languages other than English.
Write three sample conversations by hand: a short one, a typical one and a long one. Count their characters, divide by four, and feed the numbers into the calculator. Then run the same conversations against the API on a trial key and compare your estimate with the usage field in each response. If the real numbers are within 20% of your guess, the budget is trustworthy. If not, trust the real numbers and adjust.
A worked summary-memory example
Take the 40-turn chat that keeps the last 10 turns and adds a summary. Every 10 turns you send the dropped turns plus the old summary and ask for a new 100-word summary, roughly 130 tokens. That is one extra call of about 1,900 input tokens (the old summary plus 10 turns of 160 tokens, plus the instruction) and 130 output tokens, which costs 1,900 x $0.15 / 1M = $0.000285 and 130 x $0.50 / 1M = $0.000065, so $0.00035. Do it three times in the chat and the extra spend is $0.00105, while the summary adds 130 tokens to each later request. The chat still lands near $0.014, against $0.0232 for the full history, and it remembers details from turn 5 at turn 38.
Try it before you pay
New users can claim $0.50 of trial credit with one click on the API key page after sign-up. It needs no payment, is limited to one per person and is valid for 7 days. At the 20-turn size above, $0.50 covers about 73 full conversations, which is plenty to test prompts and measure real token counts. Trial keys allow 2 concurrent requests and 60 requests per minute, so run tests in a single thread. The free trial FAQ answers common questions about it, and you can claim a key on the key page.
When you are ready to build, the docs list the endpoints, streaming, tool calling and limits, and the page on a hosted uncensored LLM explains what model sits behind the API. It is one open-weight model with a 100,000-token context window, text only, and tuned to answer without content refusals for lawful adult use.
More guides
Questions and answers
How much does one chatbot conversation cost?
With a 300-token system prompt, 40-token user messages and 120-token replies, a 10-turn chat costs about $0.0022 and a 20-turn chat about $0.0068 at $0.15 and $0.50 per 1M tokens.
Why does cost grow faster than the number of turns?
Every request resends the earlier messages, so the input grows with each turn. Total input rises roughly with the square of the number of turns unless you cap history.
Are failed requests charged?
No. Failed or refused requests are not charged, and credit never expires.
How do I pay?
Top up with USDT on TRC20 or USDC on Base, any whole amount from $20 to $500. Bonuses start at $50.
Your key is one form away
Create an account, copy the key, change the base URL. That is the whole setup.