Function Compute:Billing overview

Trial quota

Function Compute provides a trial quota (a CU trial resource plan) for first-time users. If you have not purchased other types of resource plans, any usage beyond the trial quota in each billing cycle is charged on a pay-as-you-go basis.

Pay-as-you-go

You are charged based on the computing resources you actually consume. No charges apply when no resources are consumed.

Resource plans

Function Compute offers CU resource plans in monthly and yearly tiers. After you purchase a resource plan, it offsets your resource consumption first. When the plan is exhausted, billing automatically switches to pay-as-you-go. Resource plans offer lower per-unit pricing, which reduces your overall costs.

Resident resource pools (subscription)

A resident resource pool reserves GPU computing capacity through a monthly subscription. Purchase a pool, assign resident instances to your GPU functions, and get predictable costs with zero cold starts.

Billable items

Function Compute charges consist of two parts:

• CU usage: Function Compute uses CU usage as the unified billable item. All resource usage is converted to CU usage by using fixed conversion factors. This topic focuses on how CU usage is metered and billed.

• Outbound Internet traffic: Outbound Internet traffic is aggregated with other Alibaba Cloud Internet-facing products and billed on the CDT side. For more information, see CDT billing.

Unit prices

CU usage is billed on a monthly tiered basis and calculated independently by region. CU usage in each region is accumulated separately and not merged across regions. For more information, see Billing examples.

Billing cycle

Function Compute bills on an hourly basis. CU usage per function per hour is rounded up, then summed across all functions to produce the total hourly CU usage. If you have a resource plan, it offsets CU usage first. After the plan is exhausted, pay-as-you-go billing applies.

Calculation formula

All resource usage is converted to CU usage by using fixed conversion factors, then summed to get the total CU usage.

CU usage = Σ(resource usage x CU conversion factor)

Resource usage

Elastic instances have two states: active and idle. When an instance has no requests to process or no background tasks running, it automatically enters the idle state. The following table describes the scenarios for each state:

Thresholds for determining active and idle state

The system uses the following indicators to determine whether an instance has entered the idle state. All conditions must be met simultaneously:

• CPU utilization — When the instance's vCPU utilization continuously falls below 0.01 vCPU cores (equivalent to 1% utilization of a single core), the CPU condition for idle state is satisfied.

• GPU utilization — When the instance's GPU stream processors, decoders, and other components fall below the system threshold, the GPU condition for idle state is satisfied. The threshold varies by GPU series.

• Network traffic — When the instance's inbound and outbound network traffic continuously falls below 1,000 bytes/second, the network condition for idle state is satisfied.

• Request processing — The instance is not processing any HTTP requests or background tasks.

Resource usage = elastic instance (active) usage + elastic instance (idle) usage. When an instance is idle, the system automatically freezes some resources and charges at the minimum keep-alive rate:

• vCPU resources: No charge.

• GPU resources: Charged at 11.7% to 23% of the active rate depending on the GPU series:

  Ada.1 series	Ada.2 series	Ada.3 series	Tesla series	Hopper series	Hopper.1 series	Hopper.2 series	Ampere series	Blackwell.1 series	Xpu.1 series
  11.7%	11.7%	11.7%	23%	13%	14%	14%	16%	13%	19%

Setting the minimum instance count to 1 or higher is suitable for latency-sensitive workloads (session affinity, persistent connections, etc.) and background task scenarios. This provides the following benefits:

• Warm start: Achieves millisecond-level warm starts to keep services running without interruption.

• Guaranteed delivery: Locks elastic resources to guarantee resource availability.

• Smart billing for busy and idle periods: Automatically switches to idle billing when there are no requests or background tasks, which further reduces costs.

CU conversion factors