OpenClaw Pricing (2026): Free Software, Paid Runtime

OpenClaw costs $0 for the software, but a credible always-on budget is $12.48 to $24 per month in the worked VPS scenario, or $39 to $49 per month from a third-party managed provider. At $50 an hour, managed annual pricing beats DIY once maintenance consumes just 18 to 32 minutes a month.

OpenClaw pricing at a glance

OpenClaw has no official paid plan. The project charges no license fee, no seat fee, no annual subscription, and no overage because the software is MIT-licensed on GitHub. Your bill comes from the runtime around it: the computer or server, the AI model, paid tools, backups, and the person who keeps the system working.

OpenClaw documents those meters on its live API usage and costs page, which is the closest thing the project has to a pricing page.

Prices and limits were verified against OpenClaw's live cost documentation, its repository, and every named provider pricing page on August 16, 2026. The distinction between project price and operating cost matters because several companies sell managed OpenClaw hosting under their own brands. Those are third-party services, not official OpenClaw tiers.

The $12 VPS is an explicit example, not an OpenClaw minimum. DigitalOcean also lists Basic Droplets at $4, $6, $18, and $24 per month across different memory and CPU sizes. Capacity should follow the workload, not the smallest number on a hosting page.

The shortest buying rule is blunt. Reuse a machine for experimentation. Use a VPS when you need uptime and can own the server. Buy managed hosting when half an hour of monthly upkeep costs more than the hosting premium. Do not choose a path from the software's $0 price alone.

Is OpenClaw free? The free-plan truth

OpenClaw is free software, not a freemium product. There is no reduced Free edition with a Pro upgrade hiding the useful features. The same MIT-licensed codebase is available without a vendor subscription, and the project's own pages publish no Free, Pro, Team, or Enterprise ladder.

That answer covers the license and nothing else.

A genuinely $0 incremental setup is possible when all four conditions are true:

• You already own a machine that can stay on when the agent needs it.
• You run a compatible local model through Ollama.
• You keep memory embeddings local and use key-free or self-hosted search.
• You value the extra electricity, hardware wear, backup work, and maintenance time at $0 in the cash budget.

OpenClaw's Ollama provider documentation records discovered local models with a cost of $0. Local and private-network Ollama hosts also avoid a hosted model credential. That removes a provider invoice, but it does not prove the workflow is economical. The budget only wins if the outcome still passes without excessive correction time.

Who never needs to pay

A technical solo user with an existing desktop, one low-volume workflow, a local Ollama model, local memory, and no paid media or search APIs may never need to pay OpenClaw or another vendor. This is the honest free path. It is most credible for personal briefs, small document jobs, local classification, and experiments that can stop when the machine sleeps.

The free path stops when the agent must remain available away from that machine, when a local model cannot meet the acceptance bar, or when the workflow calls a metered service. A request can trigger model inference, web search, media understanding, image or video generation, speech, remote embeddings, or a skill's own API. Each meter has its own provider and billing rules.

There is no official OpenClaw student discount because there is no official paid OpenClaw plan to discount. A student can use the full open-source project under the same license as anyone else. Third-party hosting companies set their own promotions and eligibility rules.

What the monthly OpenClaw bill contains

The useful budget has five lines: host, model, tools, recovery, and labor. Leave out any one and the total will look artificially cheap.

1. Host: $0 local or a monthly server

An existing machine can make the incremental hosting line $0. The tradeoff is availability. If the laptop sleeps, loses connectivity, reboots for an update, or leaves the network, the agent is no longer an always-on service.

A virtual private server, or VPS, is a rented remote computer. DigitalOcean's live Basic Droplet page lists 512 MiB at $4/month, 1 GiB at $6, 2 GiB with 1 vCPU at $12, 2 GiB with 2 vCPUs at $18, and 4 GiB with 2 vCPUs at $24. The worked budget in this page uses the $12 option because it gives the calculation a concrete host without pretending that one server size fits every browser, channel, model, and concurrency pattern.

Server price is not model price. OpenClaw can route reasoning through an external API, leaving that usage on a separate provider bill. Verify host capacity against the actual browser, model, media, and concurrency workload before choosing the smallest plan.

2. Model: from local $0 to usage-based API billing

Every core model reply or tool turn uses the selected provider. With a direct API key, the model normally bills tokens. With a supported subscription login, the provider enforces a quota or credit window instead. With local Ollama, the catalog price is $0 and the machine supplies the compute.

The current OpenAI API prices used in the worked example are:

• GPT-5.6 Luna: $0.20 input and $1.20 output per 1 million tokens.
• GPT-5.6 Terra: $2 input and $12 output per 1 million tokens.
• GPT-5.6 Sol: $5 input and $30 output per 1 million tokens.

These are not three OpenClaw plans. They are three external model tiers routed through OpenClaw. The model can be changed without changing the software license.

3. Tools: search, media, memory, speech, and skills

The final answer is often the smallest part of an agent's action graph. One job may search the web, fetch pages, inspect an image, use remote memory embeddings, call a skill API, and then ask the core model to summarize the results. Each provider can spend a different credential.

OpenClaw's cost page gives a concrete search example. Brave Search includes $5 in renewing monthly credit, while its Search plan costs $5 per 1,000 requests. The credit therefore covers 1,000 requests a month. A workflow that makes 300 searches remains inside that credit if nothing else uses it. Another 500 requests beyond the credit have a $2.50 list-price cost.

The control is a per-workflow cap. A global monthly credit does not tell you which automation consumed it or whether retries multiplied the call count.

4. Recovery: backups and snapshots

An always-on agent stores configuration, credentials, memory, task state, and logs that can matter after a failed update. DigitalOcean prices weekly backups at 20% of the Droplet cost and daily backups at 30%, with usage-based backups starting at $0.01 per GiB per month. Snapshots cost $0.06 per GB per month.

On the $12 VPS example, percentage-based backups add $2.40 for weekly or $3.60 for daily coverage. That is small, but omitting it from every comparison makes self-hosting look cheaper precisely by excluding the thing you need when self-hosting fails.

5. Labor: setup, patches, credentials, and review

Labor is the line that flips the decision. A self-hosted runtime needs installation, network boundaries, updates, provider credentials, usage caps, logs, backup checks, and recovery. The agent's output also needs review until the workflow earns trust.

Do not assign that time a fictional universal rate. Use the value of the person who would otherwise do the work. A founder whose hour is worth $50 reaches a different answer from a hobbyist who enjoys maintaining the system. The break-even section below turns that rate into minutes, which makes the decision concrete.

Cost per accepted outcome on Luna, Terra, and Sol

Model price matters only after it is divided by work you can use. A cheap attempt that needs repair can cost more than an expensive attempt that passes.

Use one transparent workload:

• 30 attempts per month.
• 50,000 input tokens and 5,000 output tokens per attempt.
• An 80% acceptance rate, producing 24 accepted outcomes.
• A $12 VPS.
• 10 search requests per attempt, or 300 monthly, inside the Brave credit if unused elsewhere.

This is a planning scenario, not an OpenClaw benchmark. Your workflow may be much smaller or much larger. The point is to make every assumption visible so you can replace it with your own usage.

Luna: $0.52 per accepted outcome with the VPS

GPT-5.6 Luna costs $0.016 per attempt in this workload: $0.01 for 50,000 input tokens plus $0.006 for 5,000 output tokens. Thirty attempts cost $0.48. Divide that by 24 accepted outcomes and the model-only cost is $0.02 each.

Add the $12 VPS and the monthly total becomes $12.48, or $0.52 per accepted outcome. Hosting dominates this light workload. Luna is the rational default when the task is structured enough that a low-cost model still passes.

Terra: $0.70 per accepted outcome with the VPS

GPT-5.6 Terra costs $0.16 per attempt: $0.10 for input and $0.06 for output. Thirty attempts cost $4.80. The model-only price is $0.20 per accepted outcome.

With the VPS, the total becomes $16.80, or $0.70 per accepted outcome. Terra costs $4.32 more per month than Luna in this workload. At a $50 hourly labor value, Terra pays for itself if its better output saves more than 5.2 minutes of monthly review or correction.

Sol: $1 per accepted outcome with the VPS

GPT-5.6 Sol costs $0.40 per attempt: $0.25 for input and $0.15 for output. Thirty attempts cost $12. The model-only price is $0.50 per accepted outcome.

With the VPS, the total becomes $24, or $1 per accepted outcome. Sol costs $7.20 more per month than Terra here. At $50 an hour, it wins economically if it removes more than 8.6 minutes of monthly correction compared with Terra.

The correct tier is therefore not the cheapest token rate. It is the cheapest tier that clears your acceptance test after correction time. Start with Luna for bounded extraction or classification. Move to Terra when five minutes of monthly repair disappears. Move to Sol only when the harder reasoning saves at least another nine minutes or protects an outcome worth more than the price gap.

Two pricing details can break the calculation at scale. GPT-5.6 prompts above 272,000 input tokens are priced at 2x input and 1.5x output for the full request. Cache writes cost 1.25x the uncached input rate. Long histories and repeatedly rewritten prompt prefixes can therefore change the bill faster than the headline per-token rate suggests.

Managed OpenClaw pricing and the annual-lock decision

getopenclaw sells a third-party managed service for $49 month to month or $468 per year, equal to $39 per month. It is not an official paid tier from the OpenClaw project. The project's own openclaw.ai site and cost documentation do not list that plan.

The managed page says hosting, setup, model routing, updates, support, and API costs are included. It also advertises a 14-day money-back guarantee and says there is no permanent free cloud tier.

The yearly price saves $120 against 12 monthly payments: $468 instead of $588. That discount has a clean crossover. At $49 month to month, you spend less than the annual prepayment for the first nine months. Month 10 is the first whole month when $468 is cheaper. If you are still evaluating whether OpenClaw belongs in the workflow, monthly billing preserves more option value despite the higher rate.

The managed break-even is measured in minutes

Compare the managed annual equivalent with the worked $12 VPS budget:

• Against Luna's $12.48 DIY total, the $39 managed price adds $26.52. At $50 an hour, it breaks even after 31.8 minutes of monthly maintenance.
• Against Terra's $16.80 DIY total, the premium is $22.20. It breaks even after 26.6 minutes.
• Against Sol's $24 DIY total, the premium is $15. It breaks even after 18 minutes.

The $49 monthly plan moves those break-even points to 43.8 minutes for Luna, 38.6 minutes for Terra, and 30 minutes for Sol.

That makes the verdict surprisingly strict. If you spend even one short maintenance session each month on updates, credentials, uptime, backups, or troubleshooting, managed annual pricing can be economically cheaper for a $50-per-hour operator. If you already run servers and the additional upkeep is near zero, DIY retains the advantage.

The missing number on the managed page

The managed page claims that all API costs are included, but it does not publish a numeric message allowance, token allowance, credit quantity, fair-use threshold, or overage schedule. That omission prevents a buyer from calculating where a heavy workflow is throttled, upgraded, or refused.

Ask for the supported model list, concurrency, rate limits, storage, backup policy, data export, cancellation terms, and the exact rule that governs unusually high model or media usage before prepaying for a year. A flat rate is valuable only when the quantity behind it is clear enough for your workload.

At 24 accepted outcomes per month, the annual-equivalent managed price is $1.63 per accepted outcome and the monthly price is $2.04. Those numbers are predictable in the worked scenario, but they remain conditional on the provider's unpublished service limits.

API pricing versus a ChatGPT subscription

OpenClaw supports two separate OpenAI money paths: direct Platform API billing and ChatGPT/Codex subscription authentication. They can route to the same model family, but they do not share a balance.

With a direct API key, the Luna, Terra, and Sol token prices above apply. OpenClaw can estimate cost when usage metadata and local prices are available, while the provider invoice remains the billing authority.

With ChatGPT/Codex OAuth, OpenClaw uses the subscription plan's quota and credit windows. The official provider documentation says OpenClaw keeps subscription quota and Platform API billing distinct. A ChatGPT subscription is therefore not a prepaid API wallet.

ChatGPT Plus costs $20 per month and includes expanded Codex usage, with limits.

Yes, you can use a supported ChatGPT/Codex subscription with OpenClaw. The official sign-in paths are openclaw onboard --auth-choice openai or openclaw models auth login --provider openai. OpenClaw then shows the subscription plan, quota windows, and credit balance rather than pretending every token has the direct API price.

The $20 subscription can be the more predictable choice when you already pay for Plus and the OpenClaw workload fits inside its available quota. Its incremental cash cost may be $0 for that month, but its allocated cost is still part of the $20 subscription and the quota can bind. The API is better when you need clean per-workflow attribution, provider spend controls, or capacity that should not compete with your interactive subscription use.

The practical rule is simple:

• Use subscription OAuth when you want one predictable personal plan and can tolerate quota windows.
• Use an API key when each workflow needs measurable usage, separate limits, and a provider invoice.
• Keep an API-key fallback only if an unattended workflow is valuable enough to justify automatic usage charges when subscription capacity runs out.

Do not label subscription usage "free API." It is app-plan capacity being used through a supported external workflow.

Hidden OpenClaw costs to budget before launch

The expensive surprises are not the OpenClaw license. They are multipliers attached to a workload that looked small during setup.

Background calls and retries

Every model-backed reply or tool turn can consume quota or tokens. A workflow that retries after a failed fetch, repeats a search, compacts a long session, or calls a fallback model can spend several times before it returns one accepted outcome. Count attempts, not just final messages.

Set a maximum number of model turns, searches, and retries for each recurring job. A scheduled task should stop with a visible failure instead of spending until it happens to succeed.

Search and media meters

Brave's 1,000-request credit is generous for one modest workflow and easy to exhaust across several agents. Image understanding, generation, video, transcription, speech, and remote embeddings use other providers. A model budget does not cap those tools.

Create separate credentials or provider-side limits for high-risk meters where possible. The goal is attribution: which accepted outcome consumed the charge?

Backups, snapshots, and duplicate environments

A $12 server becomes $14.40 with weekly percentage-based backup or $15.60 with daily backup before tax. Snapshots, retained storage, and a second environment add more. A staging environment may be worth the money when an update can break a business workflow, but it belongs in the operating budget.

Long context and cache writes

The current GPT-5.6 price multiplier above 272,000 input tokens applies to the full request, not only the excess. Cache writes also cost more than uncached input. Persistent agents naturally accumulate history, tool results, and memory. Context discipline is therefore a pricing control, not only a speed optimization.

The invoice gap

OpenClaw's Control UI and usage commands are useful operating views, but the project says their totals describe available local session history, not the provider invoice or lifetime spend. Missing pricing metadata can leave gaps. Subscription paths may show tokens or quota without a compatible dollar estimate.

Reconcile model, host, search, media, and skill providers against their own bills. A local dashboard can be internally consistent and still omit a charge made outside its metadata.

Operator correction time

The model tier with the lowest token bill can create the highest total cost if a person repairs every result. Track review minutes beside accepted outcomes. That one column determines whether the 5.2-minute Luna-to-Terra break-even or the 8.6-minute Terra-to-Sol break-even has been crossed.

OpenClaw versus ChatGPT and n8n on price

OpenClaw is cheapest when ownership is useful. ChatGPT is cheaper in attention when you need a managed assistant. n8n is cheaper in control when the workflow should follow the same route every time.

ChatGPT: $20 per user-month for managed assistance

ChatGPT Plus costs $20 per month. It removes the VPS, operating-system updates, gateway upkeep, and separate runtime administration. It does not give you the same self-hosted control layer, but that difference only matters when the job genuinely needs it.

Read the detailed ChatGPT pricing decision if your recurring work is research, files, drafting, coding, or scheduled tasks inside a managed product. Paying $12.48 for a DIY OpenClaw stack is not a saving if maintaining it consumes an hour that ChatGPT avoids.

Choose ChatGPT when management is the feature. Choose OpenClaw when your own machine, channels, provider routing, and runtime are part of the value.

n8n: $0.008 per included execution at full Starter use

n8n prices its hosted Starter plan at $20 per month billed annually for 2,500 workflow executions, unlimited steps, unlimited users, one shared project, five concurrent executions, and 2,300 AI credits per month.

At full use, the base plan is $0.008 per included execution. That is not comparable to OpenClaw's $0.52 to $1 per accepted outcome without examining the job. An n8n execution follows a defined graph. An OpenClaw outcome can include model judgment, tool selection, and correction risk.

Use the n8n pricing breakdown when the work is a stable sequence of triggers, conditions, transformations, and destinations. Putting an autonomous agent around a deterministic webhook chain adds model spend and review without adding useful judgment.

The normalized decision is about the unit you are buying:

• OpenClaw: one operated agent runtime plus variable provider usage.
• ChatGPT Plus: one managed assistant seat-month for $20.
• n8n Starter: 2,500 included hosted workflow executions for $20 per month billed annually.

Pick the unit that matches the work. A lower price attached to the wrong operating model is not a bargain.

Which OpenClaw pricing path should you choose?

Choose the local $0 path when you already own the hardware, a local model passes the task, downtime is acceptable, and you enjoy the maintenance. It is the right place to prove one workflow before buying uptime.

Choose a self-hosted VPS when the agent must remain available, a technical owner can maintain it, and direct provider billing gives you the attribution you need. Start with the smallest server that safely supports the job, add recovery, and let measurements justify upgrades.

Choose third-party managed hosting when the setup must work this week and 18 to 32 minutes of monthly maintenance is worth more than the annual-plan premium at your labor rate. Stay monthly until the workload, provider limits, data path, and support quality survive a meaningful trial. Move annual only when month 10 looks likely.

Choose ChatGPT Plus instead when the desired outcome fits a managed assistant and nobody should own a server. Choose n8n instead when every run should follow an inspectable, deterministic path.

The broader product and security decision sits in the OpenClaw review. The pricing call is narrower: pay for OpenClaw's runtime only when runtime ownership improves the outcome.

The Monday move

Do not begin next week by choosing a plan. Begin with one accepted outcome.

1. 1

   Name one bounded job

   Choose a repeatable task with a visible finish, such as checking three official pricing pages and returning a cited change report. State what counts as accepted before the first run.

2. 2

   Run it for seven days

   Record attempts, input tokens, output tokens, searches, other paid tool calls, accepted outcomes, failures, and human correction minutes. Keep the model and workflow stable long enough for the denominator to mean something.

3. 3

   Set the provider cap

   Put a spend or quota limit on every paid provider the workflow can reach. Cap retries and background frequency inside the workflow as well.

4. 4

   Price the accepted outcome

   Add host, model, tools, recovery, and labor. Divide by accepted outcomes. Compare that number with local, VPS, managed, ChatGPT, and n8n paths using the same job.

The plan should follow that ledger. Luna, Terra, Sol, a subscription, or managed hosting earns the budget only when it lowers cost per accepted outcome or removes operational work worth more than its premium.

Frequently asked questions

How much will OpenClaw cost?

OpenClaw's software costs $0. In the worked always-on scenario, a $12 VPS plus model usage totals $12.48 per month on GPT-5.6 Luna, $16.80 on Terra, or $24 on Sol before backups, tax, other paid tools, and labor. A third-party managed provider advertises $49 monthly or $468 yearly, equal to $39 per month.

Is OpenClaw free?

Yes, the official OpenClaw project is MIT-licensed and has no paid vendor tier. Running it can still cost money for a server, model or subscription, search, media, backups, electricity, and maintenance.

What is the cheapest way to run OpenClaw?

Reuse an existing machine, run a compatible local model through Ollama, keep memory local, and avoid paid tool APIs. That can keep the incremental software, hosting, model, and search-subscription bill at $0, while electricity, hardware wear, and your time remain.

Can I use ChatGPT Plus with OpenClaw?

Yes. OpenClaw officially supports ChatGPT/Codex subscription OAuth. ChatGPT Plus costs $20 per month and its limits apply. Subscription quota is separate from direct OpenAI API billing, so Plus is not $20 of API credit.

Does OpenClaw offer a student discount?

The official project does not need a student discount because the full MIT-licensed software already costs $0. The live getopenclaw managed pricing page did not advertise a student discount when verified on August 16, 2026.

What is the OpenClaw refund policy?

The official project has no paid tier to refund. The third-party getopenclaw managed service advertises a 14-day money-back guarantee. Confirm its exact eligibility and cancellation terms with that provider before purchasing.

Has OpenClaw's price changed in 2026?

The official project's license price remains $0. Managed hosting prices come from separate providers and should not be treated as OpenClaw project price history. The getopenclaw offer verified on August 16, 2026 was $49 monthly or $468 yearly.

Want a faster way to match each AI tool to a business outcome? Get the AI Tools Map for Business Owners.