The first question almost everyone asks about OpenClaw is: "It's self-hosted, so what does it actually cost?"
Fair question. The honest answer is: probably less than you're paying for commercial AI subscriptions right now — and with far more capability. But "it depends" is not a useful answer, so this post breaks it down with real numbers.
OpenClaw costs have two components: your server (VPS) and your AI API usage. Let's cover each, then add them together.
Component 1: The VPS (Your Server)
OpenClaw runs on any Linux server. The minimum workable spec is 1 vCPU and 1 GB RAM, though 2 GB RAM is more comfortable if you run background tasks and integrations alongside it.
Current prices from major providers for a 2 GB / 1 vCPU server:
| Provider | Spec | Monthly Cost | Notes |
|---|---|---|---|
| Hetzner Cloud (CX22) | 2 vCPU / 4 GB | ~$4.50 | Best value, EU/US zones |
| DigitalOcean (Basic) | 1 vCPU / 2 GB | $12 | Simple UI, reliable |
| Vultr (Cloud Compute) | 1 vCPU / 2 GB | $10 | Many global locations |
| Linode / Akamai | 1 vCPU / 2 GB | $12 | Good uptime, solid support |
| OVH (Starter) | 2 vCPU / 2 GB | ~$5 | European alternative |
For most OpenClaw users, Hetzner is the default recommendation: a CX22 instance at roughly $4.50 per month gives you 2 vCPUs and 4 GB RAM, which is more than enough headroom. For a full walkthrough of setup, see the guide on installing OpenClaw on a VPS.
VPS cost: $5 to $12/month depending on provider.
Component 2: AI API Costs
This is the variable part, and it scales with how much you actually use your assistant. OpenClaw calls an AI model API for every conversation turn, every cron job, every automated task. You pay per token.
Here are the current prices for the most popular models (per million tokens, input/output averaged):
| Model | Input (per M tokens) | Output (per M tokens) | Good For |
|---|---|---|---|
| Claude 3.5 Haiku | $0.80 | $4.00 | Everyday use, crons |
| GPT-4o Mini | $0.15 | $0.60 | High-volume automation |
| Claude Sonnet 4 | $3.00 | $15.00 | Complex reasoning, code |
| GPT-4o | $2.50 | $10.00 | General power use |
| Gemini 1.5 Flash | $0.075 | $0.30 | Cheapest capable model |
| Ollama (local) | $0 | $0 | Full privacy, zero API cost |
Real usage numbers: a typical conversation turn with a 500-word system prompt and a 200-word reply uses roughly 400 input tokens and 200 output tokens. That's about 600 tokens per exchange.
Real-World Monthly Cost Scenarios
Let's translate that into actual monthly spend for three different usage profiles:
Light User (casual assistant, 10-15 conversations/day)
- ~400 conversation turns/month
- Model: Claude 3.5 Haiku or GPT-4o Mini
- Estimated API cost: $1 to $3/month
- VPS (Hetzner): $4.50/month
- Total: $5.50 to $7.50/month
Regular User (daily work assistant, cron jobs, 30-50 conversations/day)
- ~1,200 conversation turns/month + 200 cron job runs
- Mix: Claude Sonnet for complex tasks, Haiku for crons
- Estimated API cost: $8 to $18/month
- VPS: $5 to $12/month
- Total: $13 to $30/month
Power User (heavy automation, integrations, multiple channels)
- ~3,000+ turns/month, multiple cron agents, Telegram + Discord + n8n pipelines
- Mix: Claude Sonnet main model, GPT-4o Mini for bulk tasks
- Estimated API cost: $25 to $60/month
- VPS (larger instance for integrations): $12 to $20/month
- Total: $37 to $80/month
How OpenClaw Compares to Commercial AI Subscriptions
For context, here is what the closed alternatives cost in 2026:
| Product | Monthly Cost | API Access? | Automation? | Self-Hosted? |
|---|---|---|---|---|
| ChatGPT Plus | $20 | Limited | No | No |
| Claude Pro | $20 | No | No | No |
| Gemini Advanced | $20 | No | No | No |
| Copilot Pro | $20 | No | No | No |
| OpenClaw (light) | $6 to $8 | Full | Full | Yes |
| OpenClaw (regular) | $13 to $30 | Full | Full | Yes |
Light users pay a third of what ChatGPT Plus costs. Regular users land around the same price point but get cron jobs, n8n integration, memory, multiple channels (Telegram, Discord, WhatsApp), and full programmatic control. Power users pay more, but they're running something commercial AI products simply cannot replicate.
Three Ways to Keep Costs Low
1. Route Cheap Tasks to Cheap Models
OpenClaw lets you assign a different model to each cron job. Your morning weather check does not need Claude Sonnet. GPT-4o Mini at $0.15/M input tokens will do just fine. Reserve the expensive model for tasks that need genuine reasoning.
# cron config example
- name: daily-summary
model: google/gemini-flash-1.5 # $0.075/M input
schedule: 0 8 * * *
- name: weekly-strategy-brief
model: anthropic/claude-sonnet-4 # pays for itself
schedule: 0 9 * * 12. Use Ollama for Private or Bulk Tasks
If your VPS has spare RAM (or you run OpenClaw on a local machine), Ollama lets you run models like Llama 3.1 or Gemma 2 at zero cost. Batch document processing, private data analysis, or high-volume formatting tasks are all good candidates for local inference. No API key required, no per-token billing.
3. Use OpenRouter's Free Tier for Testing
OpenRouter gives limited free access to several models. If you're still evaluating OpenClaw and not yet in production, configure openrouter/ as your provider and experiment without spending anything.
What You Do Not Pay For
With OpenClaw, there is no per-seat licensing, no usage tier that caps your messages, no paywall on advanced features. Everything in the open-source release is yours: memory, cron scheduling, multi-channel support, skills, MCP tools, and the full agentic workflow system.
Compare that to commercial agent platforms where advanced automation, API access, or team features are locked behind $50-200/month enterprise tiers. That delta adds up fast.
Hidden Costs Worth Knowing
A few things that occasionally surprise new users:
- Context window usage: Long conversation histories accumulate tokens. OpenClaw's memory system summarises and trims context to control this, but be aware that large SOUL.md or MEMORY.md files are included in every request.
- Cron job frequency: Running a cron every 5 minutes that calls the API adds up. Use heartbeats for frequent checks and cron for scheduled tasks that actually need API calls.
- Outbound bandwidth: Minimal and usually included in VPS pricing, but worth knowing if you process large files or run heavy n8n pipelines on the same server.
Bottom Line
For most people who currently pay $20/month for ChatGPT Plus or Claude Pro and use it conversationally, OpenClaw will cost less while doing more. For automation-focused users who want cron jobs, integrations, and multi-channel access, OpenClaw is simply a different category of tool that commercial subscriptions cannot match at any price.
The realistic sweet spot for a solo professional running OpenClaw as a genuine AI assistant: $15 to $25/month total (Hetzner VPS + Claude Haiku or Sonnet API). That is comparable to one SaaS subscription and gives you a persistent, programmable, memory-enabled assistant that works across Telegram, Discord, cron automation, and everything you connect to it.
Summary
- VPS: $5 to $12/month (Hetzner CX22 is the best value pick)
- API light use: $1 to $3/month (Haiku or GPT-4o Mini)
- API regular use: $8 to $18/month (mixed models)
- API power use: $25 to $60/month (Sonnet + integrations)
- Total light user: ~$6 to $8/month (vs $20 for ChatGPT Plus)
- Total regular user: $13 to $30/month (with full automation)
- Use cheap models for crons, Ollama for zero-cost private tasks
- No per-seat fees, no usage caps, no feature paywalls
Ready to Self-Host Your AI Assistant?
Install OpenClaw on a $5/month Hetzner server and connect it to your preferred AI model. Full setup takes under 30 minutes.
Install OpenClaw Free →