Automated AI trading bot with ChatGPT, what one month of API calls costs: $9.50 to $1,512 from the public price lists

2026-10-26

43,200. That is how many calls pile up when a bot asks an AI for its next trade once a minute for 30 days (assumed). If an automated AI trading bot with ChatGPT or a similar model really makes those calls, every one of them is billed.

We worked out what that month costs from the public API price lists of three companies. The program's price, the server and the electricity are covered in other posts, so this one looks only at the model calls.

1.The unit that gets billed is the token

Token: the unit a language model counts text in. Rates are per million tokens, with input and output priced separately.

We checked the three price lists on 24 October 2026. At every company the output rate is higher than the input rate, and the top-row models show a five-fold gap, such as $10 for input and $50 for output.

Three assumptions go into the calculation.

• Call interval: once a minute, 24 hours a day, 30 days (assumed)
• Call size: 2,000 input tokens and 300 output tokens (assumed)
• Exchange rate: 1 USD = 1,400 KRW (assumed)

A month is 60 × 24 × 30 = 43,200 calls. Input tokens come to 43,200 × 2,000 = 86.4 million, output tokens to 43,200 × 300 = 12.96 million.

The monthly cost is 86.4 × input rate + 12.96 × output rate. Read two rates off a price list and any model goes through the same formula.

2.One month, worked out from three price lists

We started with the model on the top row of each list.

• OpenAI gpt-6-astra: $10 in, $50 out → 864 + 648 = $1,512.00
• Anthropic Fable 5.1: $10 in, $50 out → 864 + 648 = $1,512.00
• Google Gemini 4 Argon: $2 in, $10 out → 172.80 + 129.60 = $302.40

At 1,400 won to the dollar, $1,512 is about 2,116,800 won. $302.40 is about 423,360 won.

Now the cheapest text model on each list.

• OpenAI gpt-5-nano: $0.05 in, $0.40 out → 4.32 + 5.18 = $9.50
• Anthropic Haiku 5.5: $0.10 in, $0.50 out → 8.64 + 6.48 = $15.12 (rate for prompts of 100K tokens or less)
• Google Flash-Lite 2.5: $0.10 in, $0.40 out → 8.64 + 5.18 = $13.82

Same number of calls, different model: $9.50 against $1,512 a month, a gap of 159 times. In won that is roughly 13,300 against 2.1 million.

The three cheap models sit within $5 of each other. What moves the bill is the model tier more than the company.

Stretch the interval to five minutes

Calls drop from 43,200 to 8,640 and the bill drops to one fifth.

• gpt-6-astra, Fable 5.1: $1,512.00 → $302.40
• Gemini 4 Argon: $302.40 → $60.48
• gpt-5-nano: $9.50 → $1.90

How far one changed assumption moves the number

Raise the output from 300 to 1,000 tokens and the output charge for a top-row model jumps from $648 to $2,160, taking the month to $3,024. Output costs more than input, so the longer the answers, the steeper the bill.

Google's pricing page says the Gemini 4 Argon rate rises to $4 in and $20 out from 1 December 2026. From then on the same assumptions give $604.80.

!Rates change by prompt-length tier and by company. Re-read the price page on the day you calculate.

3.What this calculation leaves out

Model calls are billed on top of exchange fees, market data and the server. Server costs are worked out in the trading bot server post, and the price of the program itself in the trading bot price post.

The call charge is the same whether or not the model's answer was right. Nothing on any price list mentions a hit rate.

If a seller offers an AI trading bot on a monthly subscription that costs less than the figures above, it is fair to ask which model it calls and how often. Without an answer there is no way to know the AI calls exist at all.

4.One-line conclusion

If an AI trading bot really calls a large language model, the bill is calls × tokens × rate, so the public price lists alone put a month anywhere from $9.50 to $1,512, and nowhere in that number is a hit rate.

Hiddenpick is a Windows program that keeps on its server the betting rules that pick owners do not make public, and runs them as they are on the PC of the person who received them through a referral code. It currently runs on Evolution live baccarat at Stake (stake.com). How a fixed rule runs on the PC with no language model calls at all is described in the baccarat betting bot guide.

We are recruiting national pickster representatives, and those selected receive benefits worth $12,000, a one-year pass. Apply: https://hiddenpick.my/pickster

Start your 3-hour free trial