OpenClaw is open source and you do not pay to use it. What you pay for, besides the server, are the calls to the AI model: every time the agent thinks, it sends text to a provider, and the provider charges for what it processes. That is the running cost that grows with use, and it is that one, not the VPS, that tends to surprise people.
We do not write model prices here, because they change and differ from provider to provider. The current price is always on the pricing page of the provider of the model you choose (OpenAI, Anthropic, Google or another). The server price is on the VPS page.
The two separate bills
| Bill |
Who charges |
How it behaves |
| Server (VPS) |
Interweb, on a normal invoice |
Predictable: the same whether the agent is busy or idle. |
| AI model |
The model provider, through the API key |
Variable: it rises when the agent works more, reads more text or runs scheduled tasks. |
| Extras (voice, images, web search) |
The respective providers |
Each connected service has its own bill. |
What makes the model bill go up
| 1 |
Long conversations. As a rule the history is sent to the model again with every request. A long conversation costs more with every reply. See how tokens are counted.
|
|
| 2 |
Big files and pages. If the agent reads a whole document to answer a short question, you pay for the whole document.
|
|
| 3 |
Scheduled tasks. A task that runs by itself every hour spends even when you are not looking.
|
|
| 4 |
Bigger models. The more capable models usually cost more per request than the small ones. Use the big one where quality matters and the small one for the rest.
|
|
| 5 |
Retry loops. An agent that fails and retries without stopping can spend a lot before anyone notices.
|
|
How to keep the bill under control
| 1 |
Set a limit at the provider. Most let you define a spending cap or usage alerts in the console. Look for that option before you switch the agent on.
|
|
| 2 |
One key per agent. That way you see what each one spends and you can revoke one without stopping the others.
|
|
| 3 |
Check usage after a week. The provider’s console shows use per day. If there is a spike, look for the task that caused it.
|
|
| 4 |
Choose the model for the job. Summarising and sorting rarely need the most expensive model.
|
|
|
A leaked key is an open tab. Whoever holds it spends on your account. Never put it in code visible on a website or in a public repository. See how to keep a key safe.
|
|
If you want a more predictable cost, you can run models on your own VPS. In return you need a much bigger machine and more upkeep. See when it makes sense.
|
|
Need to compare server plans for the agent? See the specifications and prices on the VPS page.
See VPS plans
|
RECOMMENDED PRODUCT Web hosting with cPanel Domain and SSL included, daily backups and the panel you already know. from $5.36/mo (3-year plan, with coupon) See plans |
Português
English
