OpenClaw pricing: software, models, hosting, and tool costs

OpenClaw's software is free; the assistant you run with it may have several bills. The official project is MIT licensed and states that it has no paid tier. Your monthly total comes from model access, the machine running the gateway, and any paid services your tools call. A hosted OpenClaw offer from another company has its own price and terms.

That distinction is the starting point for a useful budget. “OpenClaw costs $X per month” is incomplete unless it names the hosting service, model access, included usage, and overage rules. The official OpenClaw overview describes the core project as self-hosted software.

The four parts of an OpenClaw bill

An existing computer may require no new monthly hosting payment, but it still needs to stay available for the jobs you expect it to run. A hosted plan may bundle infrastructure and support while excluding model tokens. Read the included services before comparing two headline prices.

Estimate model cost from the whole task

For a metered API, the basic calculation is uncached input tokens times the input rate, plus output tokens times the output rate, plus any separately priced cache reads or writes. Rates are commonly displayed per million tokens. Use the selected provider's definitions so cached tokens are not accidentally counted twice.

One message can trigger several model calls: the assistant reads context, chooses a tool, receives its result, and continues. A long answer is only part of the bill. Include the repeated context and tool results across all calls in the task. OpenClaw's token-use reference explains what contributes to usage and how estimates depend on configured rates.

For a transparent arithmetic example, assume a month contains 10 million uncached input tokens and 1 million output tokens. Kunavo currently lists Claude Sonnet 5 at $2.00 per million input tokens and $10.00 per million output tokens. Applying those rates gives a catalog estimate of $30.00 for those two categories.

Those volumes are an illustrative budget, not a measured OpenClaw workload. They exclude cache charges, hosting, and tools, and quoting a model rate does not establish an OpenClaw integration. Replace the assumed totals with your own recorded usage before choosing a spending limit.

When the upstream reports its charge, Kunavo's billing rule uses the greater of the catalog-based amount and the upstream cost multiplied by the applicable markup; otherwise chat usage bills at catalog cost. The calculation above is therefore not a guaranteed maximum charge. Reconcile your trial with the actual usage ledger. The smallest wallet top-up is $10; that funds future calls and is neither a subscription fee nor the price of one task.

Background work belongs in the budget

Budget for the assistant's schedule, not just your chat activity. A routine that checks a source every hour runs many more times than one you invoke twice a week. The relevant question is how often useful work needs to happen and how much context each run needs.

OpenClaw's API cost reference also identifies paid surfaces beyond the main conversation, including configured search, media services, memory embeddings, and third-party tools. Some have local alternatives; others have their own plan or allowance. Record the provider behind each enabled capability.

When a bill rises, separate additional completed work from wasted repetition. Review retries, duplicate schedules, unnecessarily broad context, and calls to expensive tools. Change one cause at a time so you can see whether the task still succeeds and whether the recorded charge improves.

How to see what your installation spends

• /status shows the current session and available usage information.
• /usage tokens displays token usage; /usage full can include cost when the required metadata and pricing are available.
• /usage cost gives a local cost summary.
• The Control UI's Usage view breaks available history down by dimensions such as model, provider, agent, and channel.

Use this information to find the activity responsible for spending, then reconcile with provider records over the same date range. OpenClaw's local history is not a lifetime account invoice. Missing pricing is a gap to investigate, not a zero-cost result.

Self-hosted or managed: what are you buying?

Choose self-hosting when you are comfortable maintaining the gateway and its environment. Consider a managed service when operating that environment is the burden you want someone else to handle. For any third-party offer, check whether the price includes model usage, backups, upgrades, storage, and support, and how you export your data.

Do not compare a hosting-only fee with a plan that includes model allowance as if they were the same product. Also check what happens when an allowance runs out: a pause, a rate limit, or a separate usage charge leads to a different working experience.

To evaluate the model portion, compare OpenClaw API providers, payment costs, and model rates, then follow the Kunavo custom-provider setup. The setup is based on documented configuration; verify the selected route with a bounded task before moving scheduled work. If you want to change the assistant workflow itself, use the alternatives comparison instead of treating a provider price as a reason to migrate everything.

FAQ

Official OpenClaw documentation checked September 17, 2026. Kunavo model rates and the illustrative calculation use the live site catalog.