AI Units

Overview

AI Units is the measure used to license AI products. AI Units are charged based on consumption when the models are bringing value to you.

For more information, check the Metering and charging logic and License tracking sections below.

You can also allocate and track AI Units consumption at tenant level. See the tenant-level allocation pages for more details:

Automation Cloud™ - Automation Cloud - Allocating licenses to tenants
Automation Suite - Automation Suite - Allocating robot and service licenses to tenants

Note: By default, each tenant is allocated 0 AI units, and all AI units are consumed from the account pool. If there are no AI units allocated to the tenant, AI units are consumed from the organization account pool. In case all AI Units are consumed from the tenant pool, then the admin needs to allocate more on that specific tenant.

Metering and Charging Logic

General Logic

This page contains specific information regarding AI Units depending on the used activity, covering the cost for every AI product.

To calculate the overall consumption cost, the following formula is used:

prediction cost + hardware cost = consumption cost

For more information, check the following sections below:

Prediction cost
Hardware cost

Prediction Cost

To calculate the prediction cost, the following formula is used:

input size x unit cost of the model = prediction cost

For example, the consumption for UiPath Custom Named Entity Recognition model with input size 5000 characters is as follows:

5000 characters = 3 units Consumption: 3 units x 0.5 (unit cost) = 1.5 AI Units

Input size

Model	Input type	Input size	Computed input size
Document Understanding (UiPath and Customer-Managed third party)	Document	1 page	Number of pages in the input document
Communications Mining	JSON	1 message	Number of messages per mailbox or ticketing system
Computer Vision	Image	1 image	Always 1
Task Mining	Dataset	1 dataset	Always 1
Other models	JSON	2000 characters = 1 unit	Ceil(length(input)/2000)
	File	5 MB = 1 unit	Ceil(size/5MB)
	Files	5 MB = 1 unit	Ceil(sum(size(input))/5MB)

Model used

Model	When we charge	Unit cost
Document Understanding (UiPath and Customer-Managed third party)	Per prediction	For a list of all Document Understanding models, check the Metering & Charging Logic page from the Document Understanding guide.
Computer Vision	Per prediction	0
Models in preview (like UiPath Image Classification)	Per prediction	0
Task Mining	Per successful pipeline	5000
Communications Mining	Per message uploaded, modified, or predicted	1 - for more information on Communications Mining charging logic, check the official documentation.
UiPath Light Text classifier	Per prediction	0.2
UiPath Multilingual classifier	Per prediction	0.5
UiPath Custom Named Entity Recognition	Per prediction	0.5
Open Source packages	Per prediction	0.1

Note: Except for Task Mining, running a pipeline or deploying an ML Skill only consumes AI Units related to hardware consumption.

Hardware Cost

The hardware cost at the time of deploying ML Skills is calculated as follows:

replicas x resource cost

The default replica count depends on the account type:

Enterprise account: 2
Other account types: 1

Note: Availability is increased by increasing the number of replicas. High Availability (HA) is not guaranteed if user reduces the number of replicas to 1.

Use the following table to check the resource cost for ML Skills.

Hardware	Unit Cost
0.5 CPU 2 GB RAM (default)	1 AI Unit / replica / hour
1 CPU 4 GB RAM	2 AI Units / replica / hour
2 CPU 8 GB RAM	4 AI Units / replica / hour
4 CPU 16 GB RAM	8 AI Units / replica / hour
6 CPU 24 GB RAM	12 AI Units / replica / hour
GPU	20 AI Units / replica / hour

For hardware cost related to Pipelines, check the following table.

Hardware	Unit Cost
CPU	6 AI Units / hour
GPU	20 AI Units / hour

Note: Any hour started is charged.

Consumption Example

Background Information

To automate a given process, you need to use the two following UiPath models:

The first step is to train the Multilingual Text Classification model on your dataset. The training takes 6 hours and 30 minutes using GPU.

After deploying both models as HA skills, they are running on CPU for three months. During this time, the Multilingual Text Classification model processed 20,000 texts, all around 3,000 characters, while the Invoices model processed 10,000 invoices containing 2 pages each.

Calculating Total Consumption

AI Units consumed for training Multilingual Text Classification:
7 (hours) x 20 (AI units per hour for GPU) = 140 AI Units
AI Units consumed for hosting Multilingual Text Classification for three months:
24 (hours in day) x 90 (number of days) x 2 (AI units per hour) = 4320 AI Units
AI Units consumed for hosting Invoices for three months:
24 (hours in day) x 90 (number of days) x 2 (AI units per hour) = 4320 AI Units
AI Units consumed for predictions made using Multilingual Text Classification:
20000 (number of predictions) x 2 (input size) x 0.5 (unit cost) = 20000 AI Units
AI Units consumed for predictions made using Invoices:
10000 (number of predictions) x 2 (input size) x 1 (unit cost) = 20000 AI Units
AI Units consumed in total:
hardware cost + predictions cost = (140 + 4320 + 4320) + (20000 + 20000) = 48780 AI Units