This page explains pricing for edition, compute resources, cluster operation mode, cluster management fees, and applicable ingress fees in Google Kubernetes Engine (GKE).

GKE includes fully automated cluster lifecycle management, pod and cluster autoscaling, cost visibility, automated infrastructure cost optimization, and multi-cluster management features, like Fleets, Teams, Config Management, Policy Controller, and Fleet dashboard, at no extra cost.

GKE also includes a Service Level Agreement (SLA) that's financially backed providing the following availability:

• 99.95% for the control plane of Autopilot clusters and regional Standard clusters. 
• 99.5% for the control plane of zonal Standard clusters.
• 99.9% for Autopilot Pods in multiple zones.

A flat cluster management fee of $0.10 per cluster per hour (charged in 1 second increments) applies to all GKE clusters irrespective of the mode of operation, cluster size, or topology. Whether it is a single-zone cluster, multi-zonal cluster, regional or Autopilot cluster, they all accrue the same flat fee per cluster.

For clusters in the Extended release channel, an additional cluster management fee applies if the minor version of the cluster is in the extended support period.

The following example demonstrates how the cluster management fee and free tier credit is applied for an organization’s billing accounts. In this example, the organization’s regional and zonal cluster hours are listed and the total billable amount is calculated per month, with the monthly free tier credit applied.

For Standard node pools and compute classes that don't use the Autopilot setting, you're billed for the nodes' underlying Compute Engine instances according to Compute Engine's pricing, until the nodes are deleted. Compute Engine resources are billed on a per-second basis with a one-minute minimum usage cost.

Compute Engine offers committed use discounts that apply to the Compute Engine instances in the cluster. To learn more, see committed use discounts in Compute Engine.

Autopilot clusters accrue a flat fee of $0.10/hour for each cluster after the free tier, in addition to charges for the workloads that you run. Autopilot uses a workload-driven provisioning model, where resources are provisioned based on the requirements specified in the Pod specification of your workloads. GKE includes a Service Level Agreement (SLA) that's financially backed providing availability of 99.95% for the control plane, and 99.9% for Autopilot Pods in multiple zones, or 99.99% for Autopilot Pods in multiple regions. Committed use discounts (CUDs) can be used to reduce costs for workloads that have predictable resource usage.

By default, workloads that you create are provisioned on our general-purpose computing platform where you're billed only for the resources that Pods request (and not for spare compute capacity or system overhead). If your workloads need to scale beyond 28 vCPU, you can use the Balanced or Scale-Out compute classes, which use the same approach of Pod-based compute provisioning and billing. If you enable Confidential GKE Nodes, additional charges apply. For more information, see Confidential GKE Nodes on GKE Autopilot pricing.

You can also request specific hardware like accelerators or Compute Engine machine series for your workloads. For these specialized workloads, Autopilot provides nodes that have at least the requested compute capacity for the workload and bills you for the entire node. This node-based compute model is ideal for workloads that have specific hardware requirements, but requires you to consider how to fully utilize the resources provisioned.

These compute provisioning and billing approaches mean that you can use specific compute hardware for specialized workloads while using the simpler Pod-based compute provisioning approach for everything else.

The default general-purpose platform and the Balanced and Scale-Out compute classes use a Pod-based billing model. You are charged in one-second increments for the CPU, memory, and ephemeral storage resources that your running Pods request in the Pod resource requests, with no minimum duration. This billing model applies to the default general-purpose platform and to the Balanced and Scale-Out compute classes. This model has the following considerations:

• Autopilot sets a default value if no resource request was defined, and scales up values that don't meet the required minimums or CPU-to-memory ratio. Set the resource requests to what your workloads require to get the most optimal price.
• You are only billed for Pods that are being created or are currently running (those in the Running phase, and those with the ContainerCreating status in the Pending phase). Pods waiting to be scheduled, and those that have terminated (like Pods marked as Succeeded or Failed) are not billed.
• You are not charged for system Pods, operating system overhead, unallocated space, or unscheduled Pods. Set appropriate resource requests and Pod replica counts for your workloads for optimal cost. With the Pod-based billing model, the underlying node size or quantity doesn't matter for billing.

Note: Prior to version 1.29.2-gke.1060000, and for clusters that were originally created on a version earlier than 1.26, Autopilot sets the resources requested to equal the resource limits values in cases where only the limit is set.

Note: GPU Pods that run on GKE version 1.29.4-gke.1427000 and earlier are billed according to the Autopilot GPU Pod SKUs and use the Pod-based billing model.

Some features are available via individual SKUs. These include Managed Service Mesh, Binary Authorization, Advanced Vulnerability Insights, Multicluster Gateway and Multicluster Ingress, Backup for GKE, and GKE Extended Support.

Multi Cluster Gateway only charges Pods which are direct backends of Gateway resources based on a multi-cluster gateway class. Multi Cluster Ingress only charges Pods which are direct backends of MultiClusterIngress resources. Pods which are not Multi Cluster Gateway or Multi Cluster Ingress backends are not charged. In this example, there are three MultiClusterService resources across two clusters with Pod backends. Pods that are members of Service A, B, and C are not direct backends and are not billed against the standalone pricing.

Any Pod that is a member of multiple MultiClusterService resources is billed for each MultiClusterService that they are a member of. Two Pods are members of both the D and E MultiClusterService.

The following table summarizes the monthly cost total for standalone billing for the two clusters in the example:

As an example, a customer with a single backup plan in Iowa (us-central1) that backs up an average of 20 pods during a month, storing 200GiB of backup storage data in Iowa, would be charged $25.60 in fees. This $25.60 would include $20 for the month for GKE backup management (20 x $1.00 / pod-month) and $5.60 for backup storage (200 * $0.028 / GiB-month).

Starting June 26 2023, new network outbound data transfer charges will be introduced for backups that are stored in a different region from their source GKE cluster. These charges will be based on the source and destination region and the number of bytes transferred for each such "cross-region" backup operation:

GKE cluster location