How Much Does OpenClaw Cost? A Real Pricing Breakdown
How Much Does OpenClaw Cost? A Real Pricing Breakdown
Most OpenClaw cost guides give you ranges. “$6-13 per month.” “Around $20-50 depending on usage.” Useful as starting points, but they don’t tell you where the money actually goes.
I run Klaus, so I see the line items. The AWS invoice for every instance we provision. What OpenRouter charges per token across dozens of models. How fast Orthogonal credits burn when an agent starts calling Apollo or Hunter.io. This post breaks down those numbers.
OpenClaw itself is free. It is MIT-licensed open-source software with no licensing fees (OpenClaw License). Running it costs $6 to $200+ per month depending on your server, AI model, and usage volume. Most people pay $20 to $50 per month.
Here is how those costs actually break down.
What Does OpenClaw Actually Cost? The Three Buckets
OpenClaw cost is the sum of three components: server hosting, AI model token usage, and tool or integration credits.
Most guides quote a single range and call it done. That’s like quoting “car costs” without separating the price, gas, and insurance. Each bucket has different dynamics, and you control each one independently.
Hosting is your fixed baseline. OpenClaw needs a server running around the clock. This ranges from $0 (run locally or use Oracle Cloud free tier) to $200+ per month depending on provider and specs. Pick a plan and this number stays predictable month to month.
AI model tokens are your biggest variable cost. Every conversation, automation step, and tool call sends a request to a language model. You pay per token. The price difference between a budget model and a premium one is over 20x for the same task. This bucket is where most of your money goes and where you have the most control.
Tool and integration credits are the cost most guides skip entirely. Web search, email, lead enrichment. Each integration has its own pricing. Some are free. Some are not. And if you self-host, you are subscribing to each one separately. On a managed host like Klaus, many of these come bundled.
| Cost Bucket | What It Covers | Typical Range | How It Varies |
|---|---|---|---|
| Hosting | Server running OpenClaw 24/7 | $0-200/mo | Fixed per plan |
| AI model tokens | Every API call to language models | $1-150/mo | Scales with model choice and volume |
| Tool credits | Web search, email, scraping, APIs | $0-30/mo | Depends on which tools you use |
The rest of this post breaks down each bucket with actual numbers from our billing.
How Much Does OpenClaw Hosting Cost?
OpenClaw hosting cost ranges from $0 to $200+ per month, depending on whether you self-host on a VPS or use a managed hosting provider.
Hosting is the most predictable cost. You pick a server, you pay a monthly fee. OpenClaw needs at minimum 2 vCPUs and 2 GB of RAM to run (4 GB recommended for browser automation and concurrent skills).
Self-hosting on a VPS
The cheapest path is a virtual private server. You install OpenClaw yourself, configure integrations, and handle all maintenance.
Hetzner is the budget baseline. A CAX11 (2 vCPU, 4 GB ARM) runs about EUR 3.29/mo starting April 2026 (Hetzner Cloud). Oracle Cloud offers a free ARM instance (4 OCPU, 24 GB RAM) that works for personal use. The “always free” tier is real, but instances can get reclaimed during capacity shortages, and availability varies by region. I would not rely on it for anything business-critical.
AWS, which is what we run at Klaus, costs more. A t4g.small (2 vCPU, 2 GB RAM) runs about $12/mo on-demand in us-east-1. A t4g.xlarge (4 vCPU, 16 GB) runs about $97/mo (AWS EC2 Pricing). You pay for reliability, global availability, and the AWS ecosystem.
Managed hosting
Managed hosting means someone else handles the server, updates, security, and integrations. You skip 3-8 hours of initial setup and the ongoing maintenance that follows.
At Klaus, our tiers look like this (Klaus Pricing):
| Plan | Monthly Cost | Instance | Storage | AI Credits | Included Tools |
|---|---|---|---|---|---|
| Starter | $19/mo | t4g.small (2 vCPU, 2 GB) | 30 GB | $15 one-time | $20 Orthogonal, AgentMail |
| Plus | $49/mo | t4g.medium (2 vCPU, 4 GB) | 60 GB | $30 one-time + BYOK | $20 Orthogonal, AgentMail |
| Pro | $200/mo | t4g.xlarge (4 vCPU, 16 GB) | 100 GB | $100/mo | Priority support, AgentMail |
| Managed Rollout | $1,000+/mo | Custom | Custom | Included | Full-service deployment |
Other managed hosts exist (xCloud, MyClaw, RunMyClaw). We wrote about the differences in Aren’t All the OpenClaw Hosting Services the Same?. The short version: the real gap between managed providers is what comes preconfigured. Klaus includes Orthogonal credits (Apollo, Hunter.io, SearchAPI), AgentMail, Google Workspace integration, and browser automation out of the box. Most competitors ship a bare OpenClaw instance and leave integrations to you.
Self-hosted vs managed: the real comparison
| Factor | Self-Hosted (Hetzner) | Self-Hosted (AWS) | Klaus Starter | Klaus Pro |
|---|---|---|---|---|
| Monthly cost | ~$5 | ~$12 | $19 | $200 |
| Setup time | 3-8 hours | 3-8 hours | 5 minutes | 5 minutes |
| AI credits included | No | No | $15 one-time | $100/mo |
| Integrations preconfigured | No | No | Yes | Yes |
| Monitoring (Clawbert) | No | No | Yes | Yes |
| Maintenance | You | You | Klaus | Klaus |
Hetzner is cheaper on paper. But you are also your own sysadmin, security team, and support desk. Whether that tradeoff works depends on what your time is worth.
For more on the managed vs self-hosted decision, see our hosting comparison guide.
How Much Do AI Model Tokens Cost?
AI model token costs range from $0.20 to $15 per million tokens, making model choice the single largest variable in your OpenClaw bill.
This is the bucket where most people overspend without realizing it. The choice between a budget model and a premium one can mean a 20x difference in cost for identical tasks.
OpenClaw routes API calls through OpenRouter by default. The per-token rates match what providers charge directly. OpenRouter takes a 5.5% cut when you buy credits. You pick the model, you pay close to the provider’s rate.
The most popular models break down like this (per million tokens, March 2026):
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Tier |
|---|---|---|---|
| GPT-5.4 Nano | $0.20 | $1.25 | Budget |
| DeepSeek V3.2 | $0.26 | $0.38 | Budget |
| GPT-5.4 Mini | $0.75 | $4.50 | Mid-tier |
| Claude Haiku 4.5 | $1.00 | $5.00 | Mid-tier |
| GPT-5.4 | $2.50 | $15.00 | Premium |
| Claude Sonnet 4.6 | $3.00 | $15.00 | Premium |
Sources: Anthropic Claude Pricing, OpenRouter
What this means in real money
A typical OpenClaw interaction uses about 1,000 input tokens and 500 output tokens. Here is the monthly cost at different usage levels:
| Monthly Interactions | GPT-5.4 Nano | Claude Haiku 4.5 | Claude Sonnet 4.6 |
|---|---|---|---|
| 300 (~10/day) | $0.25 | $1.05 | $3.15 |
| 1,500 (~50/day) | $1.24 | $5.25 | $15.75 |
| 6,000 (~200/day) | $4.95 | $21.00 | $63.00 |
| 30,000 (~1,000/day) | $24.75 | $105.00 | $315.00 |
The gap between GPT-5.4 Nano and Claude Sonnet 4.6 is over 12x for the same workload. No other variable in your OpenClaw bill comes close.
Model routing cuts costs 60-80%
Nobody should run one model for everything. Route 80% of routine tasks (classification, extraction, short summaries) to budget models. Save premium models for complex reasoning. We see users cut API costs by 60-80% with this approach alone.
The hidden cost: heartbeat and context accumulation
OpenClaw runs background health checks called heartbeats. Each one consumes 8,000-15,000 tokens. With a premium model, untuned heartbeats can add $30-100/mo to your bill (WenHao Yu). On Klaus, Clawbert handles instance monitoring separately, so you can reduce heartbeat frequency or disable it entirely.
Context accumulation is the other hidden drain. Every message in your conversation history gets resent with each new request. Long sessions can balloon to 100,000+ tokens per call. Use /compact to summarize and reset context length (OpenClaw docs).
What Do Tool and Integration Credits Cost?
Tool and integration credits are the third and most overlooked cost bucket, ranging from $0 to $30+ per month depending on which services your OpenClaw agent uses.
OpenClaw can call external APIs for web search, email, lead enrichment, and scraping. Each has its own pricing. Self-hosters pay for each one separately:
| Tool | What It Does | Self-Hosted Cost | Included in Klaus? |
|---|---|---|---|
| Brave Search | Web search | Free tier: ~1,000 req/mo | Yes (OpenClaw default) |
| Apollo | Lead enrichment, company data | Starts at $49/mo | Yes (via Orthogonal) |
| Hunter.io | Email finding and verification | Starts at $34/mo | Yes (via Orthogonal) |
| AgentMail | Dedicated email inbox for agent | Varies | Yes (all plans) |
| Browser automation | Chrome control for web tasks | Free (built-in) | Yes (preconfigured) |
Klaus includes $20 in Orthogonal credits on the Starter and Plus plans (Klaus Pricing). Orthogonal bundles access to Apollo, Hunter.io, SearchAPI, Exa, and other paid APIs under a single credit system. For most users, the included credits cover early tool usage. After that, top up Orthogonal credits or bring your own API keys.
Self-hosters pay each subscription individually. Apollo alone starts at $49/mo. If you need lead enrichment, email finding, and web scraping, that is $100+ per month in tool costs before AI tokens enter the picture.
Real Monthly Cost by Usage Level
These numbers combine all three cost buckets for different types of users.
| Usage Level | AI Calls/Mo | Hosting | AI Tokens | Tools | Total |
|---|---|---|---|---|---|
| Light personal | Under 1,000 | $5-19 | $1-5 | $0-5 | $6-25 |
| Active personal | 1,000-5,000 | $19-49 | $5-25 | $5-15 | $25-75 |
| Small business | 5,000-15,000 | $49-200 | $25-75 | $15-30 | $75-250 |
| Heavy automation | 15,000+ | $200+ | $75-300+ | $30+ | $250+ |
Light personal is someone checking email summaries and asking a few research questions per day. A Hetzner VPS with GPT-5.4 Nano handles this under $10/mo. On Klaus Starter ($19/mo), the included AI credits cover first-month token costs. Think daily briefings, meeting prep, and the occasional web search.
Active personal is a founder or SDR running daily research, lead enrichment, and meeting prep. This is where most of our users land. The Plus plan ($49/mo) with BYOK for OpenRouter keeps costs predictable. Typical use: pulling company profiles through Apollo, drafting outreach emails, summarizing PDFs.
Small business means multiple workflows running in parallel: lead processing, content generation, CRM syncing, daily reports. The Pro plan ($200/mo with $100 in monthly AI credits) is built for this tier. The included credits offset a significant portion of the token spend.
Heavy automation is multi-agent orchestration, RAG pipelines, and high-volume browser tasks. Budget carefully and monitor closely. One widely discussed cautionary tale: a user hit $3,600 in a single month from unmonitored workflows running premium models around the clock. That is an extreme case, but it shows what happens without spending limits.
How to Keep OpenClaw Costs Under Control
Five strategies cover 90% of the cost management problem.
Route by model tier. Start here. Send classification, extraction, and summaries to GPT-5.4 Nano or DeepSeek V3.2. Save Claude Sonnet 4.6 or GPT-5.4 for complex reasoning. This alone cuts API costs 60-80%.
Set spending limits. OpenRouter supports hard caps and alerts at custom thresholds. Set a monthly cap with alerts at 50% and 80%. A $50/mo cap prevents runaway bills from forgotten automations.
Use /compact regularly. Long conversations accumulate context. Every message resends the full history. /compact summarizes the session and resets token usage (OpenClaw docs).
Disable heartbeat on idle instances. If your OpenClaw agent is not running active automations, heartbeat checks waste tokens. Disable them or increase the interval (WenHao Yu).
Audit automations weekly. Forgotten test workflows quietly calling APIs are one of the most common cost leaks. Clawbert flags unusual activity on Klaus instances, but you should still review what is running.
Frequently Asked Questions
Is OpenClaw free?
Yes. OpenClaw is MIT-licensed open-source software with no licensing fees or subscription costs (OpenClaw docs). You pay for the server it runs on and the AI models it calls. The software itself costs nothing to download, install, or use.
What is the cheapest way to run OpenClaw?
Oracle Cloud’s free ARM tier plus a free-tier AI model costs $0/mo for very light use. For reliable daily use, a Hetzner VPS ($5/mo) plus DeepSeek V3 ($3/mo in tokens) totals about $8/mo (Hetzner Cloud, WenHao Yu).
Is managed hosting worth the extra cost?
It depends on what your time is worth. Self-hosting takes 3-8 hours to set up and requires ongoing maintenance: updates, security patches, debugging. Klaus is ready in five minutes with preconfigured integrations, AI credits, and Clawbert monitoring. If your hourly rate exceeds $20, the setup time alone costs more than a year of Starter.
How much does Klaus cost?
Starter $19/mo, Plus $49/mo, Pro $200/mo, Managed Rollout $1,000+/mo. All plans include a dedicated VM, AgentMail, and Orthogonal credits. Full details at klausai.com/pricing. For general questions, see the Klaus FAQ.
Why are some people’s OpenClaw bills so high?
Three common causes: unmonitored automations running premium models around the clock, heartbeat background checks draining tokens with expensive models, and context accumulation in long sessions. Model routing, spending limits, and the /compact command prevent all three.
Key Takeaways
- OpenClaw is free to install (MIT license). You pay for hosting, AI model tokens, and tool integrations.
- Most users spend $20-50 per month total across all three cost buckets.
- AI model choice creates a 20x+ cost difference for the same workload. That choice drives more of your bill than hosting and tools combined.
- Budget self-hosting (Hetzner + DeepSeek V3) costs about $8/mo. Managed hosting (Klaus Starter) starts at $19/mo with AI credits included.
- Model routing (80% budget tasks, 20% premium reasoning) cuts API costs by 60-80%.
- Heartbeat checks and context accumulation are hidden costs that compound without configuration changes.
- Set spending limits in OpenRouter and audit active automations weekly to prevent surprise bills.
If you would rather skip the infrastructure setup, Klaus gives you a running OpenClaw instance in five minutes with AI credits and integrations included. See the tiers at klausai.com/pricing.
Sources
- OpenClaw. “API Usage and Costs.” Accessed March 2026.
- OpenClaw. “Token Use.” Accessed March 2026.
- OpenRouter. Model pricing. Accessed March 2026.
- Anthropic. “Claude Pricing.” Accessed March 2026.
- AWS. “EC2 On-Demand Pricing.” Accessed March 2026.
- Hetzner. “Cost-Optimized Cloud Servers.” Accessed March 2026.
- Klaus. Pricing page. Verified March 2026.
- WenHao Yu. “OpenClaw Deploy Cost Guide.” February 2026.