Calculator

How much does an AI agent cost?

Enter your real workload and see two numbers: what the agent costs you per month and what it costs you to do it by hand. The per-token prices are real and update on their own.

12 models · live OpenRouter prices · updated 2026-10-22

Scenario

Answers frequent questions over chat or WhatsApp

Costs almost nobody counts

Tokens are estimates: in English, about 4 characters equal roughly 1 token. Adjust the values with your real data so the number is useful.

Estimated monthly savings

$2,192

88% less than doing it by hand.

Tokens per task$0.0054
Tokens per month$8.10
Human review (10.0 h)$250
Setup amortized over 12 months$50.00
Monthly total with AI$308
Doing it by hand (100.0 h)$2,500
Setup pays for itself in0.3 months
Annual savings$26,303

The same scenario on every model

ModelPer taskPer month
Qwen3.7 FlashQwen$0.0001$0.11
Llama 3.1 8BMeta$0.0001$0.13
GPT-5 nanoOpenAI$0.0002$0.27
Gemini 2.5 Flash LiteGoogle$0.0002$0.36
DeepSeek ChatDeepSeek$0.0006$0.93
GPT-5 miniOpenAI$0.0009$1.35
Gemini 2.5 FlashGoogle$0.0011$1.66
Claude Haiku 4.5Anthropic$0.0027$4.05
GPT-5OpenAI$0.0045$6.75
Gemini 2.5 ProGoogle$0.0045$6.75
Claude Sonnet 5Anthropic$0.0054$8.10
Claude Opus 5Anthropic$0.0135$20.25

How to read the result

The cost per task is what really decides whether an automation makes sense. A process that runs 40 times a month can carry an expensive model without any trouble. One that runs 15,000 times a month changes completely with the model you pick, and that is where the comparison table is worth more than any general recommendation.

What this math does not include

  • The time to set up and tune the automation, which you pay once.
  • Human review at the critical points, which is worth keeping.
  • Retries and errors: budget 10 to 20 percent extra tokens.
  • The cost of the tools around it (CRM, database, hosting).

FAQ

Frequently asked questions

How much does an AI agent cost per month?+

It depends on three things: how many tasks it does, how many tokens each task uses and which model it runs on. A support bot with 1,500 conversations a month on an economy model usually costs a few dollars; the same volume on a premium model can cost tens. The calculator gives you the number with real prices.

What is a token and why is it billed this way?+

A token is the smallest unit of text the model processes. In English, about 4 characters equal roughly 1 token. You are billed per input token (what you send) and output token (what it answers), and output almost always costs more.

Where do the prices come from?+

From the public OpenRouter API, which aggregates per-token prices for hosted models. They update automatically every 24 hours.

Is the saving it shows realistic?+

It is an honest estimate of the direct cost: hours of work at your rate against the cost of tokens. It does not include the time to set up the automation or human review, which is worth keeping in critical processes. Use it to decide whether it is worth exploring, not as a promise to a client.

Is the cheapest model the best choice?+

Not always. An economy model that gets things wrong costs you more in review than you save in tokens. The comparison table is there so you can see the price jump and test the quality yourself on your specific task.