UiPath Pricing: Agentic Automation Cost, License Cost and Platform Unit Rates

The honest answer has two halves. UiPath's published answer is $25 a month, which is the starting price of the Basic tier and the only dollar figure anywhere on its pricing page. The answer that matches what companies actually sign is closer to $45,000 a year, because the tier price does not include the seats, robots or consumption that make the platform useful.

Vendr, which aggregates negotiated software contracts, reports a median UiPath spend of $45,338 per year across 42 purchases, with a low of $22,440 and a high of $223,821. That is a ten-to-one spread between the smallest and largest deals in the same dataset, which tells you the quote is built from your volume rather than from a list.

Treat the last two rows with real caution. Partner guides for the same license type currently circulate figures as far apart as $1,680 and $10,000 per unattended robot per year. A five-fold disagreement between sources is not a price, it is an absence of one, and it is the practical reason to get a written quote before building any business case on a per-robot number you found online.

One derived figure is worth carrying into a negotiation. The published entry price implies $300 a year. The Vendr median is $45,338. That is about 151 times the published number, and it is the gap you are actually negotiating.

Platform Units are UiPath's single consumable under the Unified Pricing plan. Agents, document processing, communications mining and other services all draw from the same pool, and when the pool empties those operations stop. UiPath publishes how many units each operation consumes. It does not publish what a unit costs in dollars, which is the central oddity of this pricing model: the meter is public, the rate is not.

That still makes the rate card useful, because it lets you compare the relative cost of two design choices before you have a quote. Here is the published consumption, drawn from UiPath's product documentation and checked on August 29, 2026.

Two consequences fall straight out of that table and both are worth acting on.

Classification is effectively free when it precedes extraction. Classifying a page costs 0.04 units and extraction after classification costs 0.16, which totals 0.20. Extraction without prior classification also costs 0.20, and UiPath states you never pay more than 0.2 Platform Units per page for extraction. So there is no cost argument for skipping classification, and every accuracy argument for keeping it.

Generative validation doubles the price of a document. A classified, extracted page costs 0.20 units. Turn on generative validation and the same page costs 0.40. That is a defensible spend on invoices heading into a payment run and an expensive habit on low-risk documents, so scope it per document type rather than switching it on globally.

One more billing detail worth knowing: UiPath charges for LLM calls in 64,000 input token increments. An agent that stuffs a large document into every prompt crosses into a second increment and pays twice for the same call, which is an argument for trimming context rather than passing whole files.

This is the finding that moves the most money and it is not on the pricing page at all. It sits in the agents licensing documentation.

When an agent uses a UiPath-hosted model, UiPath meters every LLM call. When an agent uses a customer-managed model, meaning you supply your own model endpoint and key, UiPath charges once per agent run at the same 0.2 unit rate, no matter how many calls the agent makes inside that run.

Agents rarely make one model call. A tool-using agent typically plans, calls a tool, reads the result, decides again, and answers, which is four or five calls for one completed task. Complex agents make many more. Here is what that does to the meter.

The catch is that a customer-managed model means you now pay your model provider directly, so the saving is not free. What you are really doing is moving the marginal cost of thinking off the UiPath meter and onto a token bill you can shop around, cache and rate-limit yourself. For a high-volume agent that is usually the cheaper side of the trade, and it is the reason bring-your-own-model sits in the Enterprise tier rather than lower down.

Model choice is the second lever, and it is smaller but free to pull. A premium model costs 0.4 units per call and a basic model costs 0.16, a 2.5-fold difference for identical run counts. Most production agents do routing, extraction and form-filling work that a basic model handles, so reserving premium models for the steps that genuinely need reasoning cuts the meter without changing the architecture.

If you are comparing this against other vendors, the same structural question decides the bill on every platform. Microsoft meters Copilot Studio in Copilot Credits, Salesforce meters Agentforce in Flex Credits at 20 credits per action, and SAP meters Joule in AI Units. Our comparison of eleven AI agent platforms puts all of those meters side by side.

UiPath sells named licenses separately from the tier, and they are purchasable across all three tiers. UiPath does not publish a price for any of them. What it does publish is what each one is allowed to do, which is enough to size a quote request accurately.

The distinction that costs the most money is Pro User against Unattended Robot. A Pro User license lets a person build and run automations while they are present. An Unattended Robot is the thing that runs at 2am with nobody watching, and it is the license most commonly reported in the $8,000 to $10,000 a year band. If your business case rests on overnight processing, the robot count drives your quote far more than the developer count does.

UiPath footnotes the tier comparison with three qualifiers that are easy to miss and each one adds cost: "Users and Robots may require additional license purchases," "Some capabilities included, usage may require additional purchase," and "Limited consumption available in Basic Tier." Read together, they mean the Basic tier's $25 buys access rather than capacity.

There is a free UiPath, and whether you may legally run your business on it now depends on one number: your annual revenue. UiPath offers a free trial and a Community edition, and most guidance published online about the Community license is out of date.

Here is what the current agreement actually says. Under the Community Agreement dated July 27, 2026, clause 2.1 splits customers in two. An individual, or an organization that with its Affiliates has "less than the equivalent of USD 5 (five) million in annual revenues," is deemed a "Small Business" and "may use the Technology for internal commercial business purposes." An organization above that line is a "Large Business" and "may only use the Technology for non-commercial purposes such as testing, education, hackathons, individual or institutional research, internal trainings."

That is more permissive than the version it replaced, which is why so much published advice has it wrong. The superseded April 5, 2024 agreement licensed Community "solely for non-commercial purposes, such as education and individual" research, with no revenue carve-out at all. The 2026 terms restored a commercial allowance for smaller companies, and they also dropped the seat test that an even earlier 2022 version applied, which had required fewer than 250 physical or virtual machines or users on top of the $5 million revenue limit. Today the test is revenue only, so a 40-person US company under $5 million can staff Community as widely as it likes. Both agreements are published on UiPath's legal terms page, so check the date on any figure you find elsewhere.

Permission is not the binding constraint, though. Throughput is. Community organizations carry a hard consumption ceiling on agents of 250 LLM calls per day on UiPath-managed models, or 350 per day on customer-managed models, and that allowance resets daily for the whole organization rather than per user. So a qualifying small business is licensed to run Community commercially and still cannot process a day of invoices with it. That ceiling, not the agreement, is what pushes real workloads onto a paid tier.

The practical read for a US buyer: use Community and the trial to prove the automation works and to measure how many model calls a real run makes, then use that measured number to size the paid quote. Walking into a sales conversation with a real call-count per run is the difference between a defensible consumption estimate and accepting the vendor's.

UiPath runs two commercial licensing plans and they meter the same work differently. Unified Pricing pools everything into Platform Units. Flex keeps separate consumables: AI Units, Agent Units, Robot Units and API calls, each with its own balance.

The ratios are identical, which is the point: Flex uses a unit five times larger and the relative cost of every choice is unchanged. What differs is what happens when one kind of work runs hot. Under Unified Pricing, a document processing spike drains the same pool your agents draw from, so a busy month in accounts payable can starve a customer service agent. Under Flex, those balances are separate, so a spike is contained but you have to forecast each one correctly in advance and unused balance in the wrong bucket helps nobody.

Pick Unified Pricing when your workload mix is uncertain or shifting, which describes most first-year programs. Pick Flex when you have a stable, well-measured split between document work, robot work and agent work, and you would rather cap each one than let them compete. The choice is worth raising in the quote conversation rather than accepting whichever plan appears on the paperwork.

The tier decides capability and hosting, not headcount. These are UiPath's own descriptions, condensed, from the tier comparison on its pricing page as of August 29, 2026.

Two rows deserve attention from any US buyer. The hosting row is the one that quietly rules Basic out for regulated work: the cheapest tier is hosted in UiPath's European region and region choice begins at Standard. And the bring-your-own-model row is the reason the largest cost lever on this page, the per-run rather than per-call meter, is only available at Enterprise. If your agent volume is high enough for that lever to matter, the Enterprise tier may pay for itself through consumption alone.

UiPath is unusual in the agent platform market for publishing so little. Most of its direct competitors publish either everything or nothing, and UiPath sits awkwardly in between with one number.

The comparison that matters most to a US buyer weighing UiPath against Microsoft is the unattended runtime, because that is where RPA money goes. Power Automate Process is $150 per bot per month, published, which is $1,800 a year. Reseller guides put a UiPath unattended robot somewhere between $1,680 and $10,000 a year, unpublished. On the low estimate they are comparable; on the high estimate UiPath costs more than five times as much per runtime. That range is exactly why the quote conversation matters here and does not at Microsoft.

UiPath's counter-argument is capability rather than price, and it is a real one: self-healing UI automation, process mining and simulation, and a document stack with a published per-page meter are all things Power Automate handles less completely. Our side by side of UiPath competitors and alternatives works through where each vendor genuinely wins, and Copilot Studio against Power Automate pricing covers the Microsoft side in detail.