Agent Units, AI Units, and Platform Units: what UiPath actually charges for agentic automation in 2026, and where the budget tips over.
UiPath has billed agentic automation on a consumption basis since 2026, and that is exactly where the real budget risk lies. Anyone who only looks at the robot license when it comes to UiPath Agentic Automation overlooks three new cost drivers: Agent Units, AI Units, and the Platform Units for Autopilot. This guide shows what each metric actually costs in 2026 and where in operation the bill quietly climbs.
Agent Units are the consumption unit UiPath uses to measure the execution of AI agents in production. Development and testing at design time do not consume Agent Units; only every production run is counted. For UiPath-hosted models, the fee applies per LLM call, billed in increments of 64,000 input tokens. A call with 100,000 tokens therefore triggers two charges.
Input tokens include the system prompt, conversation history, tool definitions, and user input, but not the model's response. How many Agent Units a call costs depends on the model class (source: UiPath Agents Licensing):
The design therefore decides the bill: an agent that sends five calls to a premium model per run consumes ten Agent Units. The same agent, built leanly with a basic model and two calls, comes to 1.6. In our automation projects, we calculate these units per process in advance instead of estimating them flatly. When an AI agent versus a simple AI tool is worth it at all, read more in the linked article.
AI Units are the second consumption metric and cover document processing and ML models. One page through Document Understanding costs one AI Unit; with generative validation it is two AI Units per page. The prediction models in AI Center are counted per call (source: UiPath AI Center, AI Units):
The GPU line is the silent cost trap. A model that sits ready on a GPU replica around the clock consumes 480 AI Units a day, without processing a single document. And volumes grow fast: anyone sending 20,000 multi-page invoices through Document Understanding each month quickly ends up at 120,000 AI Units or more, depending on page count.
Under Unified Pricing, Autopilot runs on Platform Units from a consumption pool that refills every month. The allowance depends on the plan and license type and reaches up to 20 times the base tier for Enterprise Pro licenses (source: UiPath Autopilot, Unified Pricing). Not everything counts against the pool: generating, correcting, or summarizing expressions in Studio remains unlimited.
According to public pricing analyses, one action in Autopilot for Everyone costs around 0.2 Platform Units. That sounds harmless, but it adds up for every user who fires up the feature daily. Important for the calculation: Platform Units, AI Units, and Agent Units are separate pots, depending on the license model.
Two license models exist side by side: Flex books AI Units and Agent Units as separate add-ons, while Unified Pricing draws everything from a shared Platform Unit pool. The surprise arises because consumption-based metrics scale with usage, not with the number of seats. A jump from ten to a hundred agent runs a day multiplies the variable share tenfold, while a pure seat license would stay stable.
In the UiPath Community Forum, the questions revolve around exactly this: are AI Center and connected GPT models included in the Enterprise contract, or are they counted extra? The answer from the community is that all integrations are billed through Platform Units, AI Units, or Agent Units, depending on the license type (source: UiPath Community Forum). The actual prices per unit are not public; they only come through sales.
Anyone who wants to keep the risk small plans a buffer of 20 to 30 percent above the estimated consumption volume and turns on monitoring of the consumption dashboards in UiPath Insights from day one. That way you see which agent draws how many units before the monthly bill shows you. If you don't want to take this step yourself, we handle the design and operation of such agents as part of our AI agent service.
No. Agent Units only apply to production runs. Testing and developing agents at design time is excluded from Agent Unit counting, so trialing an agent does not burden the running budget.
Agent Units measure the execution of AI agents and are counted per LLM call. AI Units cover document processing and ML models in AI Center, for example one AI Unit per page in Document Understanding. They are two separate metrics with separate allowances.
No. UiPath does not publish fixed prices per Agent Unit or AI Unit. The actual terms depend on the license model and volume and are negotiated individually through sales.
As a rule of thumb, plan 20 to 30 percent above the estimated consumption volume, combined with consumption monitoring from the start. Consumption-based metrics scale with usage, so a buffer makes more sense than a pinpoint estimate.
NordFlux builds digital employees for organisations: automations and AI agents that take over repetitive work. You stay in control.
Deployed but unused UiPath models keep consuming AI Units. Why this happens and how to avoid the cost trap in AI Center.
One process, one internal validation bot: which UiPath license makes sense, why the Community Edition is ruled out, and when attended is enough.
An AI agent pursues a goal independently: it decides on the steps, uses tools, and carries out the process. Definition, examples, and limits.
Agent Units, Autopilot and AI Center bill based on consumption, and without an upfront estimate, a single agent can quickly blow past the planned budget. NordFlux calculates expected costs before the rollout and sets up guardrails that prevent budget surprises.
