Billing

Billing models

The AI2T agent service for education offers two flexible billing models.

Key concepts

Agent application types:

The platform provides the following five types of agents based on their capabilities:

1. Voice Interaction Application: An agent application that processes voice or text input and generates voice output. The end-to-end workflow includes capabilities like intent recognition, instructions, plugins, and long-term/short-term memory, but does not support image input.

2. Multimodal Interaction Application: An agent application that processes voice, text, or image input and generates voice output. The end-to-end workflow includes capabilities like intent recognition, instructions, plugins, and long-term/short-term memory. The agent supports image input.

3. Text Interaction Application: An agent application that processes voice or text input and generates text output. The end-to-end workflow includes capabilities like intent recognition, instructions, plugins, and long-term/short-term memory.

4. Translation Application: An agent application that processes voice or text input through a translation model and generates voice output. The workflow includes only the translation process and does not support standard agent capabilities such as casual chat, instructions, or plugins.

5. Visual Task Application: An agent application that processes text or image input and generates text output. This agent handles complex visual tasks, such as summarizing content from or searching for information related to IPC devices.

Definition of a conversation:

A conversation is a single turn of interaction between a user and an agent, consisting of one question and one answer. The service includes capabilities such as ASR (Automatic Speech Recognition), LLM (Large Language Model) understanding and generation, TTS (Text-to-Speech), intent recognition, instruction recognition, and long-term/short-term memory. This excludes text-to-image and image-to-image generation.

Definition of an SKU:

An SKU is a stock-keeping unit that represents a specific quantity of service conversations for a particular agent application type. When making your first purchase, you can create a new SKU. This automatically generates an SKU for the selected agent application type and quantity. For subsequent purchases, you can select an existing SKU from the drop-down list to add more conversations.

Definition of a region:

A region is the location where the agent service is deployed. Currently, you can choose between Singapore and the Chinese mainland. The unit price varies by region. Note that while the billing logic is the same for both billing models described below, the unit prices differ between regions.

Model 1: Resource pool (pay-per-conversation)

How it works When you place an order, you can specify the number of conversations included for each device. When a device is activated, the corresponding number of conversations is automatically added to a shared resource pool. This resource pool can be used by all devices under your tenant. Once the resource pool is depleted, devices can no longer initiate new conversations. You must activate more devices to replenish the pool and resume service.

Formula Total purchase cost = Unit price per conversation (varies by application type) × Quantity (conversations per device) × Number of devices

Example A

You select the Multimodal Interaction Application and purchase a prepaid plan for 100 devices, with each device allocated 100 conversations.

When you activate each of these devices, 100 conversations are added to the resource pool for the Multimodal Interaction Application. This pool is shared by all devices under your tenant that use this application type. After activating 100 devices, a total of 10,000 conversations (100 devices × 100 conversations/device) are available in the shared pool. Some devices may use more than 100 conversations, while others may use fewer. Once all 10,000 conversations are used, you must purchase for additional devices to continue the service.

Procedure

1. Select the Agent Application Type.

2. Specify the Quantity of conversations for each device.

3. Select an existing SKU or choose Create SKU.

4. Enter the Number of Devices to purchase.

5. Click Purchase.

Model 2: License model with daily limits

How it works You are charged based on the number of licenses you purchase. For a one-year service period, each license provides a device with a daily limit of 150, 300, or 500 conversations, depending on the package purchased. The daily usage limit resets at midnight.

Formula Total purchase cost = Unit price per license (varies by application type and specification) × Number of devices

Example B

You purchase 1,000 licenses. After provisioning these licenses to 1,000 devices, each device can use the service. For one year from the activation date, each device has a daily limit of 150, 300, or 500 conversations, depending on the purchased package.

Procedure

1. Select the Agent Application Type.

2. Select the resource specification for each device.

3. Select an existing SKU or choose Create SKU.

4. Enter the Number of Devices to purchase.

5. Click Purchase.

Pricing

Model 1: Resource pool (pay-per-conversation)

Model 2: License model with daily limits