For a busy agent, the AI model can cost more each month than the server it runs on. The server is a fixed price, while the model is paid per token and grows with every task. Contabo's own OpenClaw page warns that token costs can add up fast with an autonomous agent.
The calculator below adds a server from our price records to a model API at your own usage. Under it, three example jobs show how the model bill grows with use. A dated table then shows which subscriptions you can use instead of paying per token.
Who this is for
- you run OpenClaw or Hermes Agent, or plan to, and want to know the monthly total
- you are choosing a model for your agent and want to see what each job costs
- you wonder whether your ChatGPT or Claude subscription can pay for the agent's model
Not for you if you have not chosen where to host the agent yet: start with where to host an AI agent.
Key numbers
- 3 plans ChatGPT Go, Plus and Pro can pay for AI requests in participating apps, according to OpenAI Official page,
- 5 plans Claude Free, Pro, Max, Team and Enterprise, whose sign-in Anthropic keeps for its own apps Official page,
- 8 GB memory on Contabo's smallest OpenClaw plan, whose page warns that token costs add up fast Official page,
What the bill is made of
An agent's monthly cost has three parts:
- the server: a fixed monthly price, or an hourly price for the hours it runs
- the model: a price per million tokens, the short pieces of text it reads and writes, with output tokens usually dearer than input tokens
- extras: backups where they are not included, a domain, and renewals above the first-term price
The calculator
Choose a server and a model API, then enter the tokens your agent uses in a month. The results under the calculator also show three worked examples, so the page works without JavaScript.
Monthly cost of an AI agent: server and model API
| Part | Plan and price | A month | Link |
|---|---|---|---|
| Server | Hostinger KVM 2 $8.79/month 24-month term, paid up front; Black Friday first-term price, then $14.99 checked 25 Oct 2026 | $8.79 | Visit |
| Model API | Infomaniak AI Services: mistralai/Mistral-Small-4-119B-2603 €0.20 / €0.75 per 1M tokens, input / output · pay as you go, no commitment; billed at the end of each month checked 25 Oct 2026 | €1.05 | Visit |
| Total a month in US dollars | $8.79 | ||
| Total a month in euros | €1.05 | ||
Worked out as: server, 1 month at $8.79 a month; model API, 3 million input tokens at €0.20 per million plus 0.6 million output tokens at €0.75 per million.
Totals are shown per currency; we do not convert.
A month is counted as 730 hours (8,760 hours in a year divided by 12).
| Example | Server | Model API | Million tokens (in / out) | A month |
|---|---|---|---|---|
| A daily morning briefing | Hostinger KVM 2 | Infomaniak AI Services: mistralai/Mistral-Small-4-119B-2603 | 3 / 0.6 | $8.79 and €1.05 |
| Inbox sorting all day | DigitalOcean Basic Droplet, 2 vCPUs, 4 GiB | Infomaniak AI Services: mistralai/Mistral-Small-4-119B-2603 | 12 / 1.5 | $24.00 and €3.53 |
| A weekly research task | Contabo Cloud VPS 4 | Infomaniak AI Services: swiss-ai/Apertus-v1.5-70B | 6 / 1.2 | €11.60 |
The token amounts in the examples are our own examples, not measurements. Replace them with your agent's usage, which your model provider's usage page shows. The model APIs in the calculator are the ones in our price records; we add others as our weekly checks confirm them.
What makes the model bill grow
Three things drive most of the token count, and each has a setting you control:
How often the agent wakes up
An agent that checks for work every few minutes sends a prompt each time, even when there is nothing to do. Wake it on events, such as a new message, where you can.
How much it reads each time
Every request includes the instructions, the recent conversation and anything the agent has fetched. Long pages and long histories raise the input tokens of every step that follows.
Loops
An agent that retries a failing step can repeat the same expensive request many times. A spending limit on the API key stops a loop from becoming a large bill. Our guide to stopping a runaway AI bill shows which providers offer limits, prepaid credit or only alerts.
Can a subscription pay instead of tokens?
Sometimes. The providers change these rules, so check the provider's own page before you rely on it. This is what each provider's page said on 27 October 2026:
OpenAI lets Go, Plus and Pro plans pay for requests in participating apps; Anthropic keeps subscription sign-in for its own apps.
| Subscription | What the provider's page says | Source and date |
|---|---|---|
| ChatGPT Go, Plus or Pro (OpenAI) | yes, in participating apps: requests count toward your plan's Work and Codex usage, and you set a weekly limit per app | OpenAI Help Center, read 27 Oct 2026 |
| Other ChatGPT plans | sign-in only; supported open-source tools are available to all ChatGPT users | OpenAI Help Center, read 27 Oct 2026 |
| Claude Free, Pro or Max (Anthropic) | subscription sign-in is meant for ordinary use of Claude Code and Anthropic's own apps; products built by developers use API keys | Claude Code legal and compliance page, read 27 Oct 2026 |
OpenClaw's own documentation describes signing in with a ChatGPT or Codex subscription as an option. For an always-on gateway host, it calls an API key the most predictable choice. A subscription also has its own usage limits, and an agent that hits them stops until they reset.
Cutting the bill without making the agent worse
- give routine jobs a smaller, cheaper model and keep the large model for hard tasks
- wake the agent on events instead of on a short timer
- keep instructions and fetched pages short, because every step re-reads them
- set a monthly spending limit on every API key
Server or model: where to save first
Save on the model first if your agent is busy, because its cost grows with use while the server's does not. Save on the server only after you have watched its memory use for a while. An agent squeezed onto a server that is too small crashes, and a crash costs more time than the saving is worth.
Our guides to OpenClaw hosting and Hermes Agent hosting give the memory each one's documentation and the hosts suggest.
How we calculated
Prices come from our weekly price checks, each with its own date. The token volumes are our examples, not measurements. The subscription rules come from each provider's own page, read on 27 October 2026.
Mistakes to avoid
Budgeting for the server only
Instead: Add the model: for a busy agent it is the larger part.
Waking the agent every minute
Instead: Trigger it on events, or lengthen the interval.
Using a subscription where its terms do not allow it
Instead: Read the provider's own page; Anthropic's keeps sign-in for its own apps.
Running without a spending limit
Instead: Set a monthly cap on every API key.
Other questions
How much does OpenClaw cost per month?
The server costs what its price box shows, and the model adds a cost per token. The calculator above adds the two at today's prices for the volumes you enter.
Can I use my ChatGPT subscription with OpenClaw?
OpenAI's help centre says Go, Plus and Pro plans can pay for requests in participating apps. OpenClaw's documentation describes subscription sign-in; check OpenAI's list of participating apps before you rely on it.
Can I use my Claude subscription with OpenClaw?
Anthropic's Claude Code legal page keeps subscription sign-in for ordinary use of Claude Code and its own apps, and points developers' products to API keys. Use an Anthropic API key for an agent on a server.
Sources and test notes
- OpenAI Help Center: Using your ChatGPT plan in other apps and sites, read
- Claude Code documentation: legal and compliance (authentication), read
- OpenClaw documentation: model authentication, read
- OpenClaw documentation: OpenAI provider, read
- Infomaniak AI Services prices, read
- Hermes Agent home page (Nous Portal), read
- Contabo: OpenClaw hosting (token costs), read
Every page read on 27 October 2026. Token volumes are our examples; no usage of ours was measured.