The three parts of OpenClaw's cost
MIT-licensed open source. No subscription, no paid tier and no per-message fee.
Your own computer is free. A VPS keeps it online 24/7.
Usually the biggest cost. It depends on the model and how much you use it.
OpenClaw monthly cost calculator
One request that uses tools can take 3–10 model calls. Scheduled jobs count too.
Advanced assumptions
- Software$0.00
- Hosting$0
- AI model$0
This is an estimate, not a quote. It uses list prices per million tokens, ignores cache-write charges and taxes, and assumes 30 days. Agent calls are large because each one carries instructions, tool descriptions and recent history. Check your real numbers with your provider's usage dashboard. Gemini cache prices are estimated at about 10% of input.
Example monthly budgets
- Your own laptop
- GPT-6 Luna or Gemini Flash-Lite
- ~15 calls a day
- Mac mini or GPU PC at home
- Local model through Ollama or llama.cpp
- Unlimited calls, at your hardware's speed
- Hetzner VPS (~$6.49)
- Claude Sonnet 5.5 API (~$37), or a $20 plan
- ~60 calls a day
- 4 GB VPS ($24)
- Claude Opus 5.5 API (~$233), or a Max plan from $100
- ~200 calls a day, with automations
Estimates use the calculator's default assumptions (20,000 input and 600 output tokens per call, 70% cache hits). Your usage will differ.
Software: free
OpenClaw is MIT-licensed open source. There's no paid edition, no hosted tier and no per-seat or per-message fee. Every feature on this site, from chat channels to skills, is in the free download. You can use it commercially, change it and redistribute it.
The real software is free from official sources. If a site charges for OpenClaw itself, it isn't official. Get it from the official download sources.
Hosting costs
With a cloud model, OpenClaw is light: 2 GB of RAM minimum, 4 GB recommended.
| Where it runs | Specs | Monthly cost | Good for |
|---|---|---|---|
| Your own computer | What you have | $0 extra | Trying it out. It's only online while the computer is on. |
| Raspberry Pi 4/5 | 4 GB+, 64-bit | A few dollars of power | Cheapest always-on option at home |
| Hetzner CX23 | 2 vCPU · 4 GB · 40 GB | ~€5.49 / ~$6.49* | Best value VPS, with EU data centers |
| DigitalOcean Basic | 1 vCPU · 2 GB · 50 GB | $12 | Minimum spec, many regions |
| DigitalOcean Basic | 2 vCPU · 4 GB · 80 GB | $24 | Recommended spec, many regions |
| Mac mini at home | Apple silicon, 16 GB+ | Power only, after buying | Always-on with iMessage and local models |
*Hetzner raised cloud prices in 2026. Figures are from a September 2026 price tracker, so check hetzner.com. DigitalOcean figures are from its pricing page. Avoid 512 MB and 1 GB plans, which are below OpenClaw's minimum. Setup guide: host OpenClaw on a VPS.
AI model API prices
Pay-per-use prices in US dollars per million tokens. A million tokens is roughly 750,000 words. An OpenClaw agent call often uses 10,000–30,000 input tokens, because it includes instructions, tools and history.
| Model | Input | Output | Cache read |
|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | $0.25 |
| Claude Opus 5.5 | $4 | $20 | $0.20 |
| Claude Sonnet 5.5 | $2 | $10 | $0.20 |
| Claude Haiku 4.5 | $1 | $5 | — |
Source: claude.com/pricing. Cache prices from the OpenClaw Anthropic docs. Setup: connect Claude.
| Model | Input | Output | Cache read |
|---|---|---|---|
| GPT-6 Astra | $10 | $50 | $1 |
| GPT-6.1 Sol | $2 | $10 | $0.10 |
| GPT-6 Sol | $2 | $10 | $0.20 |
| GPT-6 Luna | $0.10 | $0.50 | $0.01 |
Standard rates as listed in the OpenClaw OpenAI docs. Above 272,000 input tokens, OpenAI charges more for the whole request. Check OpenAI pricing. Setup: connect OpenAI.
| Model | Input | Output | Free tier |
|---|---|---|---|
| Gemini 3.1 Pro (preview) | $2 | $12 | No |
| Gemini 3.8 Flash | $0.75* | $3.75* | Yes |
| Gemini 3.5 Flash | $1.50 | $9 | Yes |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 | Yes |
Source: Gemini API pricing. Gemini 3.1 Pro prices are for prompts up to 200K tokens. *Gemini 3.8 Flash's introductory price doubles to $1.50 / $7.50 on January 1, 2027. Setup: use Gemini.
Use a subscription you already pay for
Instead of paying per token, OpenClaw can use some consumer plans. You pay a flat fee, and usage counts against that plan's limits.
| Plan | Price (US) | How OpenClaw uses it |
|---|---|---|
| ChatGPT Plus | $20/month | Codex login: openclaw models auth login --provider openai |
| ChatGPT Pro | From $100/month | Same, with much higher limits |
| Claude Pro | $20/month ($17 billed annually) | Claude CLI route, reusing your Claude Code sign-in |
| Claude Max | From $100/month | Same, with higher limits |
| Gemini | — | Not available. Google ended Gemini CLI Google sign-in, so use an API key. |
The cost of local models
Run a model with Ollama, LM Studio or llama.cpp, and every message is free.
Small models need 8–16 GB of memory. Better ones need 24–32 GB with a GPU or Apple silicon. That's a one-time cost if you don't already own it.
A machine drawing 30 W around the clock uses about 22 kWh a month, which is roughly $4 at $0.20 per kWh. A GPU under load draws much more.
Weigh the trade-offs, such as quality on complex tasks and speed, in OpenClaw with local models.
Seven ways to cut your OpenClaw bill
- Set a spending limit in every provider dashboard before anything else.
- Use a cheaper model for routine work. Give your scheduled-jobs agent Haiku, GPT-6 Luna or Flash-Lite, and keep the expensive model for hard tasks.
- Keep prompt caching on. With an Anthropic API key, OpenClaw caches automatically. Set
cacheRetention: "long"for long sessions with pauses. - Lower the thinking level with
/think lowfor simple requests. - Avoid huge contexts. OpenAI and Gemini charge more above 272K and 200K input tokens.
- Add a local or cheap fallback so busy days don't all run on your priciest model. See model setup.
- Leave Fast mode off. Fast mode costs about 2x on Claude Opus 5.5 and OpenAI.
openclaw models set openai/gpt-6-luna --agent automationsTrack what you're spending
Add ANTHROPIC_ADMIN_KEY or OPENAI_ADMIN_KEY to see 30 days of actual cost in the Control UI Usage page.
ChatGPT Codex sign-ins show plan, quota windows and credit balance in OpenClaw.
openclaw models status shows which model and account each agent uses, so a costly default doesn't slip through.
OpenClaw pricing questions
Is OpenClaw free?
Yes. OpenClaw is free, MIT-licensed open-source software with no paid tier. You only pay for hosting if you use a server, and for the AI model unless you use a local model or an existing subscription.
How much does OpenClaw cost per month?
It ranges from $0, on your own computer with a local model, to several hundred dollars for heavy use of top models. Many personal setups cost about $5 to $50 a month: a small VPS plus a mid-priced model or a $20 plan.
What is the cheapest way to run OpenClaw?
Run it on a computer you already own with a free local model. The cheapest always-on option is a Raspberry Pi with a low-cost API model such as GPT-6 Luna or Gemini Flash-Lite.
Can I use my ChatGPT Plus or Claude Pro subscription with OpenClaw?
Yes. ChatGPT plans work through Codex login, and Claude plans work by reusing your Claude Code sign-in on the same machine. Usage counts against the plan's limits. Gemini consumer sign-in is no longer available, so use an API key.
Why is my OpenClaw API bill higher than expected?
Each agent call includes instructions, tool descriptions and conversation history, often 10,000 to 30,000 input tokens. One request can also take several calls when tools are used, and scheduled jobs add more. Use caching, a cheaper model for routine agents, and spending limits.
How much does a VPS for OpenClaw cost?
A 4 GB VPS costs about $6.49 a month at Hetzner (EU) or $24 at DigitalOcean, as of September 2026. A 2 GB server is the minimum. Avoid 512 MB and 1 GB plans.