AI agent cost calculator: what your agent costs each month
By the serverserv team · 5 min read · Updated
Most of an AI agent's cost is the model, not the server, and it depends on how often the agent calls the model and how much context it sends each time. Enter your usage below to estimate the monthly bill, then see the same workload priced on other models.
The calculator
Messages answered, emails handled, jobs run.
Each agent step or tool round trip is one call.
System prompt, memory, chat history and tool results.
The reply, tool calls and any thinking tokens.
The same workload on other models
| Model | Price in / out per 1M | Model cost per month |
|---|
How AI agent costs add up
An AI agent is not one question and one answer. To finish a task it usually calls the model several times: once to plan, once for each tool it uses, and once to write the result. Every call sends the model the whole context again, which means the system prompt, the agent's memory, the conversation so far and whatever the last tool returned.
That is why input tokens, not output, make up most of an agent's bill. A short reply can sit on top of 5,000 to 30,000 tokens of context, and the context grows as the task goes on. Thinking or reasoning tokens are billed as output, so models that think a lot cost more per call than the reply length suggests.
The calculator multiplies four numbers: tasks per day, calls per task, tokens per call and the model's price, over 30 days. Then it adds the server the agent runs on. If you do not know your token numbers, start with the defaults and adjust once you can see real usage in your provider's dashboard.
Three common examples
| Agent | Usage | Claude Haiku 5.5 | GPT-6.1 Terra | Claude Sonnet 5.5 |
|---|---|---|---|---|
| Personal assistant on WhatsApp or Telegram | 20 tasks a day, 3 calls, 5,000 in and 500 out per call | $1.35 a month | $13.50 a month | $27 a month |
| Customer support inbox agent | 150 tasks a day, 4 calls, 8,000 in and 700 out per call | $20.70 a month | $207 a month | $414 a month |
| Research or coding agent | 10 tasks a day, 25 calls, 30,000 in and 2,000 out per call | $30 a month | $300 a month | $600 a month |
Add the server on top: from $14.99 a month for OpenClaw on serverserv. The model you pick changes the bill far more than anything else, so it pays to test whether a smaller model handles your routine tasks well enough.
How to cut your AI agent bill
- Use a small model for routine work. Send simple replies, sorting and summaries to a low-cost model and keep the large model for hard tasks.
- Turn on prompt caching. Most providers charge much less for context they have seen recently, and agents resend the same system prompt on every call.
- Keep memory short. Summarise old conversation instead of sending all of it, and trim large tool results before they go back to the model.
- Cap the steps. Set a maximum number of tool calls per task so one confused run cannot loop for an hour.
- Set a spending limit in your model provider's dashboard, and check usage weekly for the first month.
Where hosting fits
The server is the small, fixed part of the cost. An agent like OpenClaw needs to stay online so it can answer messages and run scheduled jobs, which is why people run it on a VPS rather than a laptop. A 2 vCPU, 4 GB RAM server is enough when the model runs at a provider. You only need a much larger server if you want to run a local model yourself. More on this in how much OpenClaw costs and OpenClaw VPS requirements.
Questions
How much does it cost to run an AI agent per month?
For one person, often $2 to $30 a month in model usage on a small or mid-size model, plus the server. Busy agents that handle hundreds of tasks a day on a large model can cost several hundred dollars a month. The calculator above gives an estimate for your own usage.
Why is input so much more expensive than output for agents?
Each call resends the system prompt, memory, history and tool results, so an agent reads far more tokens than it writes. Prompt caching and shorter memory are the main ways to bring that down.
Does serverserv charge for AI model usage?
No. You bring your own API key and your model provider bills you directly. Your serverserv plan covers the server and the setup.
Can I run an AI agent with no model costs?
Yes, by running a local model on your own server, but you need a lot more RAM and the answers are usually slower and weaker than the large hosted models. Most people start with a hosted model and a small server.
How accurate are the model prices?
They are standard API prices from each provider's pricing page on 28 October 2026. Prices change often and some providers have scheduled changes, so check before you rely on a number.