Free calculator
What does an AI agent cost each month?
Pick an agent, how much it works and where it runs. You'll see three ways to pay for the AI, plus the time it takes you to keep it working.
How the estimate works
Prices checked on
Each total adds four things: AI use, where the agent runs, the agent software and your upkeep time (your hours times what an hour is worth). AI use is billed by the amount of text it reads and writes. An agent re-reads a lot of the same text, and vendors charge less for text the AI has already read (cached input), so that share is priced at the cached rate. Some vendors charge extra the first time text is stored for reuse; we price that at the normal input rate, so pay-as-you-go totals for those vendors may run a little higher. Totals are rounded to the nearest dollar and shown in US dollars. We check the prices every month.
Not included: taxes, buying a computer, setup work and any other tools you connect. You pay for your own AI use, directly to the vendor.
There's no AI bill: Ollama runs open models on your own computer with no subscription. You pay for the computer and its power. Small models need 8 GB of memory, and bigger, more capable ones need 32 GB or more.
How the usage levels were sized
Light: a few tasks a day: About 60 scheduled tasks and 10 longer conversations a month, sized from the agent jobs Dave Swift runs every day. Upkeep time is our estimate: in the last 30 days Hermes shipped 11 updates and OpenClaw shipped 27, so expect a little time each week.
Daily: handles part of your workday: About 10 scheduled tasks and 2 longer conversations a day, sized the same way. On Claude Sonnet 5.5, the AI use alone comes to about $40 a month, inside the $20 to $50 a month that most people running OpenClaw report, according to Kilo.
Heavy: works most of the day: About 30 scheduled tasks and 6 long conversations a day, sized the same way. More connections means more things to fix, so upkeep time goes up.
Sources
- Hermes Agent on GitHub (license and the '$5 VPS' line) —
- Hermes Agent: which computers it runs on —
- Hermes Agent: AI providers it connects to —
- Hermes Agent: running a local model with Ollama (memory needs) —
- Hermes Agent: signing in with a Grok plan —
- OpenClaw on GitHub (license) —
- OpenClaw: install and system requirements —
- OpenClaw on Raspberry Pi (minimum memory) —
- OpenClaw: signing in with an AI subscription —
- OpenClaw: AI providers it connects to —
- OpenClaw: local models —
- OpenClaw: xAI and Grok —
- ChatGPT plans and prices —
- OpenAI Help: ChatGPT Pro tiers —
- OpenAI: Sign in with ChatGPT in open-source apps —
- OpenAI: using a ChatGPT plan in open-source apps —
- OpenAI Help: ChatGPT and Codex usage limits —
- Claude plans and prices —
- Anthropic: Claude Code legal and compliance (plan sign-in rules) —
- Grok plans and prices —
- xAI: Connect Grok to Hermes Agent —
- Google AI plans (US) —
- Anthropic API prices —
- OpenAI API prices —
- Google Gemini API prices —
- xAI API prices —
- DeepSeek API prices —
- OpenRouter: DeepSeek V4.1 Flash —
- OpenRouter: Qwen3.8 Flash —
- DigitalOcean Droplet prices —
- xCloud prices (AI agent hosting) —
- Elestio: managed OpenClaw —
- Apple: Mac mini power use —
- US Energy Information Administration: average home electricity price (Aug 2026) —
- Ollama prices (local models are free) —
- Kilo: How much does OpenClaw cost? (reports from people running it) —
- Hermes Agent release history —
- OpenClaw release history —
The prices behind the estimate
| Model | Vendor | Input / 1M | Cached input / 1M | Output / 1M | Note |
|---|---|---|---|---|---|
| Claude Haiku 5.5 | Anthropic | $0.10 | $0.01 | $0.50 | This price is for prompts up to 100,000 tokens. Longer prompts cost 5 times as much, and agent prompts often run longer. |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $0.10 | $10.00 | |
| Claude Opus 5.5 | Anthropic | $4.00 | $0.20 | $20.00 | |
| GPT-6 Luna | OpenAI | $0.10 | $0.01 | $0.50 | |
| GPT-6.1 Sol | OpenAI | $2.00 | $0.10 | $10.00 | |
| GPT-6 Astra | OpenAI | $10.00 | $1.00 | $50.00 | |
| Gemini 3.8 Flash | $1.50 | $0.15 | $7.50 | Regular price. Google charges half this until December 31, 2026. | |
| Gemini 4 Argon | $4.00 | $0.20 | $20.00 | Regular price. Google charges half this until November 30, 2026. | |
| Grok 4.7 | xAI | $2.00 | $0.50 | $6.00 | This price is for prompts under 200,000 tokens. Longer prompts cost about twice as much. |
| DeepSeek V4.1 Flash (open-weight) | DeepSeek | $0.30 | $0.006 | $1.20 | DeepSeek's peak-time price. Off-peak use costs half. |
| Qwen3.8 Flash (open-weight, via OpenRouter) | Alibaba Qwen via OpenRouter | $0.15 | $0.016 | $0.47 |
| Plan | Monthly price | Works in Hermes | Works in OpenClaw | Note |
|---|---|---|---|---|
| ChatGPT Plus | $20.00 | Yes | Yes | OpenAI says Plus and Pro users can use their plan to power open-source apps they run themselves, by signing in with ChatGPT. Apps that are paid or hosted for you need OpenAI's approval. Usage limits aren't published as numbers, and from December the allowance is shared between ChatGPT and Codex. |
| ChatGPT Pro 100 | $100.00 | Yes | Yes | Same agent rules as Plus. Pro also comes as Pro 200 ($200) and Pro 500 ($500). |
| ChatGPT Pro 200 | $200.00 | Yes | Yes | Same agent rules as Plus. OpenAI doesn't publish its usage limits as numbers. |
| Claude Pro | $20.00 | Limited | Limited | Anthropic's rules don't allow a third-party agent to sign in with a Claude plan. The clean route is an Anthropic API key, which is billed separately. $17 a month if you pay yearly. |
| Claude Max | $100.00 | Limited | Limited | 5x Pro usage, plus $100 a month in API credits. Those credits can power an agent through an API key. Same sign-in rules as Pro. |
| Claude Ultra | $200.00 | Limited | Limited | 10x Pro usage every week and no 5-hour limit, plus $200 a month in API credits that can power an agent through an API key. Same sign-in rules as Pro. |
| Grok Premium | $30.00 | Yes | Unclear | xAI says connecting a Grok plan to Hermes Agent works on every tier. Hermes's docs say some accounts get blocked and have to switch to an API key. xAI hasn't said anything about OpenClaw. The usage pool refreshes weekly. |
| Grok Super | $100.00 | Yes | Unclear | xAI recommends Super for daily agent users. Same agent rules as Premium. |
| Google AI Pro | $19.99 | Unclear | Unclear | Google hasn't said whether this plan can power a third-party agent. Both agents connect to Gemini with an API key, which is billed separately. |
| Google AI Ultra 5x | $99.99 | Unclear | Unclear | Same as Google AI Pro. Ultra 20x is $199.99 a month. |
Questions
How much does OpenClaw cost per month?
The software is free and open source. You pay for the AI it uses and a place to run it, such as a computer you already own or a small cloud server. Kilo's roundup of people running OpenClaw says most land somewhere between $20 and $50 a month.
How much does Hermes Agent cost?
Hermes Agent is free and open source, and its makers say it runs on a $5 server. Most of the monthly cost is the AI: a ChatGPT or Grok plan you already pay for, pay-as-you-go use with an API key, or a free model on your own computer.
Can I use my ChatGPT subscription with an AI agent?
Yes, with Plus or Pro. OpenAI says those plans can power open-source apps you run yourself when you sign in with ChatGPT. The usage limits aren't published as numbers, and from December 2026 they're shared with ChatGPT and Codex.
Can I use my Claude subscription with an AI agent?
Not by signing in. Anthropic's rules don't allow a third-party agent to sign in with a Claude plan. The clean route is an Anthropic API key, and Claude Max and Ultra include monthly API credits you can use that way.
Is a local model free?
There's no AI bill. Ollama runs open models on your own computer with no subscription. You still pay for the computer and its power, and you need at least 8 GB of memory for small models and 32 GB or more for bigger ones.
What's the cheapest way to run an AI agent?
A free local model on a computer you already own has no AI bill, and the budget pay-as-you-go models cost little for light use. Models differ in quality and speed, so count your own upkeep time and try the setup on real work before you decide.
Do I need a server to run an AI agent?
No. Both agents run on a Mac, Windows or Linux computer that you leave on. A small cloud server keeps the agent running when your computer is off, and a managed host like Elestio runs it for you.
Rather not look after it yourself?
Server Savant's Systems plan, led by Dave Swift, includes setting up one AI agent for your business and keeping it current, and you pay for your own AI use. If you'd rather do it yourself with Dave's guidance, the AI Intensive is made for that.