All Products
Search
Document Center

Platform For AI:Elastic Algorithm Service (EAS) billing

Last Updated:Jun 16, 2026

Understand EAS billing rules for each resource type to control costs before deploying model services. Covers billable items and billing methods, including pay-as-you-go, subscription, and serverless billing.

Billing overview

Deploying services with EAS incurs charges for compute resources, system disks, and dedicated gateways:

  • Compute resources: Public resources, EAS dedicated resources, general-purpose computing resources 2.0, and Lingjun intelligent computing resources.

    General-purpose computing 1.0 is available only for DSW and DLC. General-purpose computing 2.0 is available for DSW, DLC, and EAS.
  • System disk (optional): A free quota is provided (30 GiB for public resources, 200 GiB for EAS dedicated resources and general-purpose computing resources 2.0). Usage beyond the free quota is billed separately.

  • Fully managed dedicated gateway (optional): Deployments use a shared gateway (free) by default. You can purchase a fully managed dedicated gateway and configure it manually.

EAS offers two billing modes:

  • Pay-as-you-go: Billed by resource occupancy duration, not by the number of service calls. Suitable for unpredictable or fluctuating workloads.

    Note

    For SDWebUI and ComfyUI, EAS provides serverless versions. Deployment is free. You're billed only for the actual processing time when the service handles requests.

  • Subscription: Pay upfront for a commitment period at discounted rates. Suitable for long-term, stable workloads.

image
Important

If you use other Alibaba Cloud services such as Elastic IP Address, OSS, or NAS, those services are billed separately.

Billing period

Pay-as-you-go

Monitor when billing starts and stops for pay-as-you-go resources to avoid unexpected charges. Stop unused model services to prevent unnecessary costs.

Resource type

Billing starts

Billing stops

Public resources

When the service instance provisions resources (billing continues even without service calls).

When the model service is stopped (resources are released).

Important

You must stop or delete the service instance.

EAS dedicated resources /

General-purpose computing resources 2.0

When the machine is created and enters the Running state (billing continues even without deployed services).

  • When the machine resource is deleted.

  • When the machine resource enters the Stopped state.

Important

The Stopped state is triggered only by overdue payments. You cannot manually stop machines. To stop billing, delete the pay-as-you-go machine resources.

System disk

When the system disk is purchased.

  • When the EAS dedicated resources or general-purpose computing resources 2.0 machine is deleted.

  • When the service deployed with public resources is deleted.

Dedicated gateway

When the dedicated gateway is created.

When the dedicated gateway is deleted.

Warning

Common misconceptions about stopping billing for public resources on pay-as-you-go:

  1. Billing continues during startup failures: Billing is based on underlying Pod resource occupation time. Regardless of whether the business runs, billing continues as long as resources are occupied.

  2. Stopped instances: Services stopped due to overdue payments don't automatically restart after account top-up. Manually stopped services won't restart via auto-scaling policies. To resume the instance, manually restart it from the console or through an API call.

Scaling

  • Scale-out: Billing for new resources starts when the scale-out completes.

  • Scale-in: Billing for released resources stops when the scale-in completes. Remaining resources continue to be billed.

Subscription

Resource type

Billing starts

Billing stops

EAS dedicated resources /

General-purpose computing resources 2.0

00:00:00 on the day after purchase.

At the expiration time.

AI computing resources (Lingjun)

00:00:00 on the day after purchase.

At the expiration time.

System disk

When the system disk is purchased.

At the expiration time.

Dedicated gateway

When the gateway is purchased.

At the expiration time.

Serverless services

Deploying a serverless service is free. You're billed only for the actual processing time when the service handles requests.

  • Example: You submit an image generation request in the WebUI, and the model runs for 10 seconds to produce the image. The billed duration is 10 seconds.

  • Limitation: Only SDWebUI and ComfyUI support serverless deployment. Check the deployment page for pricing details.

Billable items

The following table lists the billing method for each resource type:

  • Except for serverless services, pay-as-you-go billing is calculated at one-minute granularity.

  • If you use public resources on a pay-as-you-go basis, stop unused model services to avoid unnecessary costs.

  • Some instance types may be temporarily out of stock in certain regions and can't be purchased.

Billable item

Billing method

Billing formula

Unit price

Public resources

Pay-as-you-go

(specified machine resources)

Bill per service = Number of instances × (CPU cores × (CPU unit price / 60) + Memory (GB) × (Memory unit price / 60)) × Duration (minutes)

For pricing details, see Table 1. Specified machine resources The pricing information provided below is for reference only. For actual prices, refer to the purchase page. Resource type Price CPU 0.03 (USD/core/hour) Memory 0.004 (USD/GB/hour).

Listed prices are hourly rates for readability. Actual billing uses per-minute rates (hourly price ÷ 60).

Pay-as-you-go (specified machine type)

Bill per service = Number of instances × (instance type unit price / 60) × Duration (minutes)

Pricing varies by region and instance type. Refer to the console for actual prices.

For the list of supported instance types, see Appendix: Public resource group instance types.

EAS dedicated resources /

General-purpose computing resources 2.0

Pay-as-you-go

Bill per resource group = Number of machine types × (unit price / 60) × Duration (minutes)

For pricing details, go to the EAS dedicated machine pay-as-you-go purchase page.

Subscription

Bill per resource group = Number of machine types × Unit price × Duration (months)

For pricing details, go to the EAS dedicated machine subscription purchase page.

AI computing resources (Lingjun)

Subscription / Pay-as-you-go

Purchase Lingjun intelligent computing resources and use the corresponding resource quota to deploy EAS services.

For billing details about AI computing resources, see Billing of AI computing resources.

System disk

Pay-as-you-go

Bill = Number of instances × System disk capacity (GiB) × (unit price / 60) × Duration (minutes)

For pricing details, go to the Block Storage (ESSD Cloud Disk PL1) page.

Subscription

Bill = Number of machine types × System disk capacity (GiB) × Unit price × Duration (months)

For pricing details, go to the Block Storage (ESSD Cloud Disk PL1) page.

Dedicated gateway

Pay-as-you-go

Bill = (Gateway unit price / 60) × Number of gateway nodes × Duration (minutes)

For pricing details, go to the EAS dedicated gateway pay-as-you-go page.

Subscription

Bill = Gateway unit price × Number of gateway nodes × Duration (months)

For subscription pricing details, go to the EAS dedicated gateway subscription page.

Service inference (serverless)

Billed by actual processing time when the service handles requests

Bill = Inference duration (seconds) × Unit price (per second)

Only SDWebUI and ComfyUI support serverless deployment. Check the deployment page for pricing details.

Table 1. Specified machine resources

Important

The pricing information provided below is for reference only. For actual prices, refer to the purchase page.

Resource type

Price

CPU

0.03 (USD/core/hour)

Memory

0.004 (USD/GB/hour)

Usage notes

Pricing information is for reference only. Actual costs appear on your billing statements.

Billing examples

Important

The following billing examples are for reference only. For actual prices, refer to the purchase page.

Public resource group billing examples

Pay-as-you-go example

  • Scenario: A model service is deployed using a public resource group with specified machine resources in the China (Hangzhou) region.

    • 09:00:00 - The service enters the Running state, occupying 2 CPU cores + 8 GB memory.

    • 10:00:00 - Scale-in completes, reducing resources to 1 CPU core + 4 GB memory.

    • 11:00:00 - Scale-out completes, increasing resources to 4 CPU cores + 16 GB memory.

    • 12:00:00 - The service is stopped.

  • Cost calculation:

    Bill = (2 × 0.03 + 8 × 0.004) + (1 × 0.03 + 4 × 0.004) + (4 × 0.03 + 16 × 0.004) = 0.322 USD

EAS dedicated resource billing examples

Subscription example

  • Scenario:

    Purchase two 4-core CPU + 15 GB GPU T4 machines in the China (Hangzhou) region on subscription for 3 months. Unit price: 570 USD/month (actual prices vary).

  • Cost calculation:

  • Total = 2 × 570 × 3 = 3420 USD

Pay-as-you-go example

  • Scenario:

    Purchase two ecs.g6.6xlarge (24-core CPU + 96 GB) machines in the China (Hangzhou) region on pay-as-you-go for 45 minutes. Unit price: 1.02 USD/hour (actual prices vary).

  • Cost calculation:

  • Bill = 2 × (1.02 / 60) × 45 = 1.53 USD

System disk billing examples

Subscription example

  • Scenario: Two machine resources are purchased in the China (Hangzhou) region using EAS dedicated resources on a subscription basis, each with a 300 GiB system disk, for 3 months.

  • Cost calculation:

  • Bill = 2 × (300 - 200) × 0.153 × 3 = 91.8 USD

Pay-as-you-go example

  • EAS dedicated resources

    • Scenario: Two machine resources are purchased in the China (Hangzhou) region using EAS dedicated resources on a pay-as-you-go basis, each with a 300 GiB system disk, for 5 hours.

    • Cost calculation:

    • Bill = 2 × (300 - 200) × 0.000319 × 5 = 0.319 USD
  • Public resource group

    • Scenario: Two instance nodes are purchased in the China (Hangzhou) region using a public resource group on a pay-as-you-go basis, each with a 300 GiB system disk, for 5 hours.

    • Cost calculation:

    • Bill = 2 × (300 - 30) × 0.000319 × 5 = 0.8613 USD

FAQ

How do I view deduction items and billing details?

On the Billing Details page, you can filter and view detailed PAI bills. For details, see View PAI billing details.

How do I view pay-as-you-go billing details and the corresponding resource ID for EAS public resources?

  1. Get the resource instance ID:

    The Resource Instance ID in EAS public resource pay-as-you-go billing details uses the format <Service instance ID>;<Service ID>. Find it as follows:

    1. Log on to the PAI console and go to the Elastic Algorithm Service (EAS) page. On the Inference Service tab, find the service deployed with public resources and get the service ID.

    2. Click the service name to go to the details page. In the Overview tab, scroll to the Service Instance section at the bottom to get the service instance ID.

      If the service is stopped, you can't get the instance ID from this location.
  2. View billing details: Go to the Bill Details page. Filter by product name EAS Service Pay-As-You-Go and resource instance ID <Service instance ID>;<Service ID>.

Note

If you receive an overdue payment notification but can't find the corresponding instance in the console, check the service list across all regions (especially China (Shanghai) and China (Hangzhou)) for any remaining instances that haven't been deleted.

Does a CNY 0 order generated after activating EAS incur any charges if no instance is created?

The EAS pay-as-you-go activation order is free. A CNY 0 order only indicates that the service module is activated. Charges are incurred only when EAS instances (Pods) are actually created. If no running EAS instances appear in the console, no charges are incurred.

How is SLA compensation calculated when an EAS service is unavailable?

EAS SLA compensation is calculated on a per-service (task) basis: the compensation amount is based on the unavailability duration and all resources used by that specific service, not on the total resources under your account.

Why am I still billed after stopping an EAS service?

  • Billing delay: For DSW/EAS pay-as-you-go resources, there's a delay between usage and bill generation. The charge notification you received may be for usage before you stopped the instance, not for charges incurred after stopping.

  • EAS service still running: Not using a service doesn't mean it's stopped. As long as the service isn't manually stopped or deleted, billing continues because of resource occupation. Back up any necessary data before deleting.

Why am I charged extra when using EAS trial resources?

  • Cause: When deploying stable-diffusion-webui or comfyui in EAS, if you select an image version with the -api or -cluster suffix, these versions use an asynchronous queue resource during deployment. This resource is billed and can't be offset by EAS trial credits.

  • Solution: When deploying stable-diffusion-webui or comfyui, select the standard version, which has the -standard suffix or no suffix at all.

Appendix: Public resource group instance types

The following table lists some of the public resource instance types supported by EAS. For a complete list, go to the Resource Information section on the custom deployment configuration page. For details, see Custom deployment. Supported instance types vary slightly by region. Refer to the console for the latest availability.

CPU类型

实例规格

vCPU

内存(GB)

ecs.c7.large

2

4

ecs.c7.xlarge

4

8

ecs.c7.2xlarge

8

16

ecs.c7.4xlarge

16

32

ecs.c7.6xlarge

24

48

ecs.c7.8xlarge

32

64

ecs.c7.16xlarge

64

128

ecs.r7.4xlarge

16

128

ecs.r7.large

2

16

ecs.r7.xlarge

4

32

ecs.r7.2xlarge

8

64

ecs.r7.6xlarge

24

192

ecs.r7.8xlarge

32

256

ecs.r7.16xlarge

64

512

ecs.g7.large

2

8

ecs.g7.xlarge

4

16

ecs.g7.2xlarge

8

32

ecs.g7.4xlarge

16

64

ecs.g7.6xlarge

24

96

ecs.g7.8xlarge

32

128

ecs.g7.16xlarge

64

256

ecs.g6.large

2

8

ecs.g6.xlarge

4

16

ecs.g6.2xlarge

8

32

ecs.g6.4xlarge

16

64

ecs.g6.6xlarge

24

96

ecs.g6.8xlarge

32

128

ecs.c6.large

2

4

ecs.c6.xlarge

4

8

ecs.c6.2xlarge

8

16

ecs.c6.4xlarge

16

32

ecs.c6.6xlarge

24

48

ecs.c6.8xlarge

32

64

ecs.r6.large

2

16

ecs.r6.xlarge

4

32

ecs.r6.2xlarge

8

64

ecs.r6.4xlarge

16

128

ecs.r6.6xlarge

24

192

ecs.r6.8xlarge

32

256

ecs.g5.6xlarge

24

96

ecs.c5.6xlarge

24

48

ecs.g8y.large

2

8

ecs.g8y.xlarge

4

16

ecs.g8y.2xlarge

8

32

ecs.g8y.4xlarge

16

64

ecs.g8y.8xlarge

32

128

ecs.g8y.16xlarge

64

256

ecs.c7a.large

2

4

ecs.c7a.xlarge

4

8

ecs.c7a.2xlarge

8

16

ecs.c7a.4xlarge

16

32

ecs.c7a.8xlarge

32

64

ecs.c7a.16xlarge

64

128

ecs.g7a.large

2

8

ecs.g7a.xlarge

4

16

ecs.g7a.2xlarge

8

32

ecs.g7a.4xlarge

16

64

ecs.g7a.8xlarge

32

128

ecs.g7a.16xlarge

64

256

GPU类型

实例规格

vCPU

内存(GB)

GPU显存

ml.gu7i.c8m30.1-gu30

8

30

1 * 24 GB

ml.gu7i.c16m60.1-gu30

16

60

1 * 24 GB

ml.gu7i.c32m188.1-gu30

32

188

1 * 24 GB

ml.gu7i.c64m376.2-gu30

64

376

2 * 24 GB

ml.gu7i.c128m752.4-gu30

80

256

4 * 24 GB

ecs.gn5i-c4g1.xlarge

4

16

1 * 8 GB

ecs.gn5i-c8g1.2xlarge

8

32

1 * 8 GB

ecs.gn5-c4g1.xlarge

4

30

1 * 16 GB

ecs.gn5-c8g1.2xlarge

8

60

1 * 16 GB

ecs.gn5-c8g1.4xlarge

16

120

2 * 16 GB

ecs.gn5-c28g1.7xlarge

28

112

1 * 16 GB

ecs.vgn6i-m4-vws.xlarge

4

23

1 * 4 GB

ecs.vgn6i-m8-vws.2xlarge

10

46

1 * 8 GB

ecs.gn6i-c4g1.xlarge

4

15

1 * 16 GB

ecs.gn6i-c8g1.2xlarge

8

31

1 * 16 GB

ecs.gn6i-c16g1.4xlarge

16

62

1 * 16 GB

ecs.gn6i-c24g1.6xlarge

24

93

1 * 16 GB

ecs.gn6i-c24g1.12xlarge

48

186

2 * 16 GB

ecs.gn6i-c24g1.24xlarge

96

372

4 * 16 GB

ecs.gn7i-c8g1.2xlarge

8

30

1 * 24 GB

ecs.gn7i-c16g1.4xlarge

16

60

1 * 24 GB

ecs.gn7i-c32g1.8xlarge

32

188

1 * 24 GB

ecs.gn7i-c32g1.16xlarge

64

376

2 * 24 GB

ecs.gn7i-c32g1.32xlarge

128

752

4 * 24 GB

ecs.gn6v-c8g1.2xlarge

8

32

1 * 16 GB

ecs.gn6v-c8g1.4xlarge

16

64

2 * 16 GB

ecs.gn6v-c8g1.8xlarge

32

128

4 * 16 GB

ecs.gn6e-c12g1.3xlarge

12

92

1 * 32 GB

ecs.gn6e-c12g1.12xlarge

48

368

4 * 32 GB

ecs.gn6e-c12g1.24xlarge

96

736

8 * 32 GB

ecs.gn7-c12g1.3xlarge

12

94

1 * 40 GB

ecs.gn7-c13g1.13xlarge

52

378

4 * 40 GB

ecs.gn7-c13g1.26xlarge

104

756

8 * 40 GB

ecs.gn7-c13g1.6xlarge

26

189

2 * 40 GB

ecs.gn7e-c16g1.4xlarge

16

125

1 * 80 GB

ecs.gn7e-c16g1.8xlarge

32

250

2 * 80 GB

ecs.gn7e-c16g1.16xlarge

64

500

4 * 80 GB

ecs.gn7e-c16g1.32xlarge

128

1000

8 * 80 GB