All Products
Search
Document Center

API Gateway:Choose an instance type

Last Updated:Jun 03, 2026

API Gateway offers Serverless and Dedicated (VPC) instances. Compare their specifications to choose the right type for your workload.

1. API Gateway instance types

An API Gateway instance is a resource group that processes your APIs, including public IP addresses, internal IP addresses, outbound public IP addresses, load balancers, servers, and storage. API groups must run on a valid instance. API Gateway provides the following instance types:

1.1 Instance type description

  • Dedicated instance (VPC): Dedicated instances have the same technical specifications as shared instances (VPC), but provide a higher SLA guarantee. Purchase a higher RPS specification to use dedicated resources: inbound public IPs, VPC inbound IPs, outbound public bandwidth, and isolated server clusters. Pricing of dedicated instances.

  • Serverless instance: Serverless instances (formerly shared instances) are being upgraded region by region. Existing shared-instance workloads are unaffected. Billing and pricing remain the same. Serverless instances offer higher scalability and performance, suitable for fluctuating traffic that requires timely, maintenance-free scaling.

1.2 Technical specifications and limits

Specifications and limitations

Serverless instance

Dedicated instance (VPC)

Scenarios and suggestions

Suitable for production environments of small and medium-sized businesses.

Suitable for medium- and heavy-load production environments.

Cloud resources

Shared by all tenants. Scales out automatically based on workload.

Exclusive to a single user.

SLA

99.9%

99.95% to 99.99% depending on the instance specification.API Gateway Service Level Agreement

Billing methods

Number of API calls and network traffic.

Instance specification (pay-as-you-go or subscription) and network traffic.

Inbound public IP address

Not fixed; shared by multiple tenants.

Fixed and exclusive.

Inbound IP address in a VPC

Not fixed; shared by multiple tenants.

Supports access from all VPCs in the same region.

Fixed and exclusive. Only supports access from a specific VPC.

To allow access from multiple VPCs in the same region, submit a ticket.

Outbound public bandwidth

Shared by multiple tenants. IP address is not fixed.

Exclusive. IP address is fixed.

Inbound or outbound IPv6 traffic

Unsupported.

Supported.

Maximum timeout period at the backend

60s.

3600s.

Call by API Gateway

Unsupported.

Supported.

Limits on the request body size

8 MB.

32 MB.

Circuit breaker plug-in

Configured by default; not customizable.

Customizable.

Caching

10 MB per user.

For more information, see Dedicated instances

Plug-in metadata and quantity

Max 50 KB metadata per plug-in. Max 10,000 plug-ins per region per user.

Max 50 KB metadata per plug-in.

Max 10,000 plug-ins per region per user.

Limits on a backend routing plug-in

Max 160 parameters per plug-in.

Max 160 rules per plug-in.

Max 160 parameters per plug-in.

Max 160 rules per plug-in.

Limits on an IP address-based access control plug-in

Max 160 parameters per plug-in.

Max 160 rules per plug-in.

Max 160 parameters per plug-in.

Max 160 rules per plug-in.

Uploading of trace logs to the Managed Service for OpenTelemetry console

Supported.

Supported.

Logging of HTTP requests and responses

Supported.

Supported.

Default domain name to support access over HTTPS

Unsupported.

Supported.

Maximum number of inbound requests per second (RPS)

Upper limit for each API group: 2,500.

2,500 to 60,000. For more information, see Dedicated instances

Maximum number of inbound connections

100,000 connections per domain name.

50,000 to 1,000,000.

Maximum number of new inbound connections per second (CPS)

5,000 to 20,000 (shared by all users).

5,000 to 20,000.

Maximum size of an outbound connection pool

2000

1,200 to 28,800.

Maximum inbound public bandwidth (Mbit/s)

5,120 Mbit/s shared across all tenants; no per-user limit.

5120 M

Maximum outbound public bandwidth (Mbit/s)

Shared across all tenants. Total bandwidth = number of outbound IPs × 200 Mbit/s.

100 to 600 Mbit/s.

2. Instances supported in each region

RegionId

Region name

Serverless instance

Dedicated instance (VPC)

cn-qingdao

China (Qingdao)

Supported

Supported

cn-beijing

China (Beijing)

Supported

Supported

cn-zhangjiakou

China (Zhangjiakou)

Supported

Supported

cn-huhehaote

China (Hohhot)

Supported

Supported

cn-wulanchabu

China (Ulanqab)

Supported

Supported

cn-hangzhou

China (Hangzhou)

Supported

Supported

cn-shanghai

China (Shanghai)

Supported

Supported

cn-shenzhen

China (Shenzhen)

Supported

Supported

cn-heyuan

China (Heyuan)

Supported

Not supported

cn-guangzhou

China (Guangzhou)

Supported

Supported

cn-chengdu

China (Chengdu)

Supported

Supported

cn-hongkong

China (Hong Kong)

Supported

Supported

ap-northeast-1

Japan (Tokyo)

Supported

Supported

ap-northeast-2

South Korea (Seoul)

Supported

Supported

ap-southeast-1

Singapore

Supported

Supported

ap-southeast-3

Malaysia (Kuala Lumpur)

Supported

Supported

ap-southeast-5

Indonesia (Jakarta)

Supported

Supported

ap-southeast-6

Philippines (Manila)

Supported

Supported

ap-southeast-7

Thailand (Bangkok)

Supported

Supported

us-east-1

US (Virginia)

Supported

Supported

us-west-1

US (Silicon Valley)

Supported

Supported

eu-west-1

UK (London)

Supported

Supported

me-east-1

UAE (Dubai)

Supported

Supported

eu-central-1

Germany (Frankfurt)

Supported

Supported