- Front page
- Prices
- OpenClaw
What does OpenClaw cost? Hosting and API bills
The software costs nothing and has no paid tier, which moves the real bill to two places: the model you plug in and the machine that keeps it running.
OpenClaw is released under the MIT licence and looked after by the non-profit OpenClaw Foundation, with no subscription, no paid edition and no vendor account. Your spending goes elsewhere. Every token the agent reads or writes is billed by the model provider, via an API key or under a plan you already have, unless you run a local model. If the agent lives on a rented server rather than a machine at home, a small VPS adds somewhere between $5 and $20 a month. Our estimator below turns list API prices into a monthly figure.
Plans that include OpenClaw
| Plan | Price | What you get |
|---|---|---|
| OpenClaw (self-hosted) Full agent | Free | No subscription and no paid edition: the software is free, and your monthly bill is whatever the connected model and the machine running it happen to cost. |
Cached input is the rate to watch. An agent resends its context at every step, and on GPT-6.1 Sol, for instance, repeated context costs $0.10 per million tokens, a twentieth of the $2 fresh rate.
A loop is the classic surprise: an agent stuck repeating a step keeps spending tokens until you stop it, so watch usage closely in the first weeks.
Model choice moves the bill more than anything else. Output on GPT-6 Astra costs 100 times as much per token as on GPT-6 Luna.
A local model removes the token bill altogether and lets the agent run fully offline.
What OpenClaw costs to run
Running OpenClaw on a computer you already own adds nothing to your hosting bill. An agent meant to stay on around the clock is better placed on a small VPS or a home server, and a rented VPS typically comes to about $5 to $20 a month. The installer supports macOS, Linux, WSL2 and Windows and brings Node along if it is missing, so there is no software to buy. Whatever the machine, the gateway should listen only on loopback and be reached through SSH or Tailscale rather than an exposed port.
The larger and less predictable cost is the model. You can pay per token with an API key, use a subscription you already have, or run a local model that costs nothing per token. List prices per million tokens range from $0.10 in and $0.50 out on GPT-6 Luna to $10 and $50 on GPT-6 Astra, with Claude Sonnet 5.5 and GPT-6.1 Sol both at $2 and $10 and Claude Opus 5.5 at $4 and $20. Because an agent resends its context at every step, the cached input rate often decides the bill, and your provider must also serve your country.
Model fees, about $43.20 a month
These figures use list API prices and assume that half of each task's input tokens are served from cache at the cheaper rate and that a month has 30 days.
Questions readers ask
Is OpenClaw completely free?
The software is: it is MIT-licensed and has no paid edition. What costs money is the model behind it, billed by the token or folded into a subscription you hold, plus a rented server if the agent does not live at home.
How much does a VPS for OpenClaw cost?
Renting a small VPS so the agent is always reachable usually costs around $5 to $20 a month. Running it on a computer at home avoids that bill. Whichever you choose, keep the gateway off the open internet and connect to it through SSH or Tailscale.
Which model is cheapest to run OpenClaw with?
Of the hosted models in our table, GPT-6 Luna is by far the cheapest, at $0.10 per million input tokens and $0.50 per million output tokens. A local model costs nothing in tokens at all. Run a week of your usual tasks on your choice before deciding whether it is good enough.