0%·19 min left
Cloud Cost Optimization

GKE Pricing Guide: Autopilot vs. Standard (2026)

DevZero Team

October 26, 202619 min read
GKE Pricing Guide: Autopilot vs. Standard (2026)

Google Kubernetes Engine (GKE) is one of the most widely used managed Kubernetes services, and its pricing is easy to get wrong. There are two modes, Autopilot and Standard, and they bill compute in different ways. That difference decides how much control you have and where your money goes.

This guide walks through what you pay for on GKE, how Autopilot and Standard differ, what the discounts and add-on charges look like, and how to choose between the two modes. Worked examples use real us-central1 rates.

If you want to lower your bill, we cover that separately in our GKE cost optimization guide. This post is about reading the bill, so you know what you're looking at before you start cutting.

For official rates and the latest regional pricing, check the GKE pricing documentation.

Last updated: 2026-11-15. GCP Compute Engine on-demand, Spot, and Committed Use Discount pricing for Linux in us-central1, pulled from the DevZero instances dataset. See the full set at the instance pricing explorer.

How GKE Pricing Works#

Both modes give you a managed Kubernetes control plane. What differs is how compute is billed and who manages the nodes.

ComponentAutopilot ModeStandard Mode
Control plane$0.10/hour per cluster (same as Standard)$0.10/hour per cluster
Compute resourcesCharged per pod (requested vCPU & memory)Charged per node (GCE VM pricing)
StorageCharged separatelyCharged separately
Network egressCharged separatelyCharged separately
Load balancingCharged separatelyCharged separately
GPUs & SpotSupported (GPU via node-based billing; Spot Pods available)Full support

In Autopilot, you pay for the CPU, memory, and ephemeral storage your pods request. Google provisions and manages the nodes behind the scenes. In Standard, you manage your own node pools and pay for the VMs (Google Compute Engine pricing) whether or not your workloads use them.

The Control Plane Fee#

Both modes charge a flat $0.10 per hour per cluster. That is about $73.00/month at roughly 730 hours in an average month.

The GKE free tier credit ($74.40/month per billing account) offsets this fee for one zonal Standard or Autopilot cluster. Regional clusters and additional clusters pay the full fee. See the GKE Free Tier section below.

Where Bills Creep Up#

Each mode wastes money in its own way. In Autopilot, oversized pod requests are billed even when the pods use a fraction of them. In Standard, overprovisioned nodes leave CPU and memory idle, and you still pay for all of it.

What GCE Nodes Cost In us-central1#

The control plane fee and per-pod rates above are Google list prices as of October 8, 2026. In Standard mode, node cost is where most of the bill comes from, so the tables below show it directly.

DevZero generates these tables at build time from the same pricing data that powers our instance pricing explorer. They cover Linux in us-central1 (Iowa) and refresh whenever the site is rebuilt. The rates and totals quoted in the prose and worked examples on this page are fixed to the October 8, 2026 snapshot, so they may differ from what the tables show when you read this. Use the tables as the compute baseline for any Standard mode estimate. The EKS pricing guide and AKS pricing guide have the same breakdown for AWS and Azure node pools.

On-Demand#

Machine TypevCPURAMOn-Demand $/hrApprox $/month
e2-medium24 GiB$0.0553/hr~$40.38/mo
e2-standard-228 GiB$0.0670/hr~$48.92/mo
e2-standard-4416 GiB$0.1340/hr~$97.83/mo
n2-standard-4416 GiB$0.1942/hr~$141.80/mo
n2-standard-8832 GiB$0.3885/hr~$283.58/mo
n2-highmem-4432 GiB$0.2620/hr~$191.28/mo

On-demand is the no-commitment rate. It's a good default for spiky or short-lived clusters and a poor one for a baseline that runs 24/7.

Spot#

Machine TypeOn-DemandSpotSpot Discount
e2-medium$0.0553/hr$0.0332/hr~40%
e2-standard-2$0.0670/hr$0.0402/hr~40%
e2-standard-4$0.1340/hr$0.0804/hr~40%
n2-standard-4$0.1942/hr$0.1165/hr~40%
n2-standard-8$0.3885/hr$0.2330/hr~40%
n2-highmem-4$0.2620/hr$0.1572/hr~40%

Spot VMs use spare Compute Engine capacity at a discount, but Google can reclaim them at any time, with little or no advance notice. By default the preemption notice is 0 seconds (you can opt in to 120 seconds), and the shutdown period is best effort, up to 30 seconds (Google's Spot VM documentation, fetched October 8, 2026). They suit stateless workers, batch jobs, and CI runners behind a node pool that can tolerate interruptions. As of the October 8, 2026 snapshot, the table above shows about 40% off on-demand for these e2 and n2 SKUs. Google advertises up to 91% for many machine types, and its GKE pricing page notes that Spot prices are dynamic and can change up to once every 30 days (GKE pricing), so none of these percentages is guaranteed. The Spot column is computed by DevZero's pricing pipeline from Google's Cloud Billing Spot SKUs (a per-vCPU rate plus a per-GiB rate), and it can differ from the figures on Google's own Spot VM pricing page.

Committed Use Discounts (1-Year and 3-Year)#

Machine TypeOn-Demand1-yr CUD3-yr CUD3-yr Discount
e2-medium$0.0553/hr$0.0348/hr$0.0249/hr~55%
e2-standard-2$0.0670/hr$0.0422/hr$0.0302/hr~55%
e2-standard-4$0.1340/hr$0.0844/hr$0.0603/hr~55%
n2-standard-4$0.1942/hr$0.1224/hr$0.0874/hr~55%
n2-standard-8$0.3885/hr$0.2447/hr$0.1748/hr~55%
n2-highmem-4$0.2620/hr$0.1651/hr$0.1179/hr~55%

CUDs trade a 1- or 3-year spend commitment for a lower rate on steady-state capacity. Size the commitment to the floor of your usage, not the peak.

Autopilot Mode Pricing#

Autopilot is the fully managed mode. You define what your workloads need (CPU and memory), and Google provisions and scales the compute underneath. You're billed only for the resources your pods request, not for the VMs they run on.

What Autopilot Bills You For#

Autopilot bills four things:

  • vCPU, per core requested per second.
  • Memory, per GiB requested per second.
  • Ephemeral storage, per GiB requested per second. A container that requests none gets a 1 GiB default (Google's Autopilot resource request defaults).
  • System overhead, which Google does not bill for. Unmodified system DaemonSet pods, OS overhead, unallocated capacity, and unscheduled pods are free. You only pay for what your running pods request.

As of October 8, 2026, per Google's published GKE pricing page, these are the rates for general-purpose Autopilot workloads in us-central1:

ResourcePrice
vCPU$0.0445 per vCPU per hour
Memory$0.0049225 per GiB per hour
Ephemeral storage$0.0001389 per GiB per hour

These are Google's default (on-demand) Autopilot rates. Google also publishes Spot and committed-use-discount (CUD) rates for the same resources (see the Discounts section below). There is no fixed per-pod overhead fee.

Pod Cost Example#

Say you run a pod that requests 1 vCPU, 2 GiB of memory, and 1 GiB of ephemeral storage. Your cost per hour:

  • vCPU: $0.0445 × 1 = $0.0445
  • Memory: $0.0049225 × 2 = $0.009845
  • Storage: $0.0001389 × 1 = $0.0001389
  • Total: $0.0545/hour per pod ($39.77/month at 730 hours)

Cost scales linearly with pod count and uptime.

Where Autopilot Works Well#

  • You don't manage nodes or autoscaling. Google handles node pools, sizing, and upgrades.
  • The control plane fee is the same $0.10/hour as Standard, and the free tier covers it for one cluster.
  • Pods that are waiting to be scheduled or have terminated aren't billed.

What Autopilot Gives Up#

  • Small pods can cost more than their raw request if they're rounded up to Autopilot's resource-request minimums.
  • You get less control over node-level settings.
  • Overestimated pod requests turn directly into a higher bill.

Standard Mode Pricing#

Standard gives you full control of the cluster's infrastructure. You manage the node pools, instance types, autoscaling, and lifecycle. That buys flexibility, and it also makes cost management your job.

Unlike Autopilot, you're billed for the VMs (nodes) running your workloads, not for pod-level requests.

What Standard Bills You For#

Standard bills three things:

  • The control plane, at $0.10/hour per cluster (~$73.00/month at ~730 hours in an average month).
  • Compute, based on Google Compute Engine (GCE) pricing for the VM types you choose.
  • Storage and networking, charged separately.

You can run on any GCE instance type: general-purpose (such as e2 and n2), high-memory, high-CPU, or custom shapes. Spot and preemptible instances are available for savings, and GPU instances cover ML and AI workloads.

Node Pool Cost Example#

Take a node pool of 3× n2-standard-4 VMs (4 vCPU, 16 GiB memory each), priced from the on-demand rate in the October 8, 2026 snapshot of the live table above. The table shows $0.1942/hr, rounded. The unrounded rate is $0.19424/hr, and the totals below use it:

ItemCalculationPrice
Per node (on-demand)$0.19424 × 730$141.80/month
Total (3 nodes)$0.19424 × 3 = $0.58272/hr; × 730$425.39/month

Add $73.00/month for the control plane ($0.10/hr × 730 hours) and you get $498.39/month, plus any attached disks, snapshots, or load balancers. For a zonal cluster, the free tier credit covers the $73.00 control plane fee (see below).

Where Standard Works Well#

  • You get the full range of GCE features: custom VMs, Spot, and GPUs.
  • You can tune workloads and autoscaling to fit your traffic.
  • Cluster-level optimizations and cost-aware scheduling are available.

What Standard Asks Of You#

  • You provision, scale, and optimize the node pools yourself.
  • Overprovisioned nodes are wasted spend, since you pay for the VM either way.
  • Nodes sit underutilized unless someone watches them.

GKE Free Tier#

GKE gives you $74.40/month in credits per billing account. The credit applies to one zonal Standard or Autopilot cluster and does not apply to regional clusters. It covers the control plane fee for that cluster in full. Nothing is credited against compute (VMs or pod requests), storage, or networking.

Free Tier ComponentIncluded
Control plane (one zonal Standard or Autopilot cluster)Up to $74.40/month
Control plane for regional clusters or additional clustersNo
Compute (VMs or pods)No
Storage / Network / LBsNo

If you run a single zonal cluster, in either mode, the $0.10/hour control plane charge is fully offset. Once you add a second cluster or move to a regional one, charges start accruing.

What The Free Tier Does Not Cover#

  • Regional clusters, and any cluster beyond the first per billing account, are excluded from the credit.
  • Node usage is still billed, even in Standard.
  • Free tier credits can't be applied to any other SKU, and unused credits don't roll over (GKE pricing, fetched October 8, 2026).

For a dev or test setup, the free tier can cover the base cost of one cluster. The workloads you run on it are always billed separately.

Charges Outside Compute#

Beyond compute and the control plane fee, GKE bills for the services around it. These tend to go unnoticed until the invoice arrives.

Persistent Storage#

  • Persistent storage is billed separately through Compute Engine Persistent Disk.
  • Standard PD costs $0.04/GiB/month and SSD PD costs $0.17/GiB/month (us-central1, as of October 8, 2026; see Persistent Disk pricing).
  • Standard snapshots are incremental. You're billed for the compressed size of the data that changed since the previous snapshot (same page).

Network Egress#

Egress is charged by destination and traffic volume (as of October 8, 2026, per VPC network pricing):

  • Traffic to the same zone over internal IPs in the same VPC network is free.
  • Traffic to a different zone in the same region is $0.01/GiB.
  • Internet egress to North America: the first 1 GiB per month is free, then $0.12/GiB up to 1,024 GiB, $0.11/GiB up to 10,240 GiB, and $0.08/GiB above that. Rates differ by destination and Network Service Tier.

Load Balancers#

Google Cloud Load Balancers are billed on forwarding rules and on data processed. Forwarding rules cost $0.025/hour in total for the first 5 rules and $0.01/hour for each additional rule, plus per-GiB data processing charges (see Cloud Load Balancing pricing, fetched October 8, 2026).

Backups and Snapshots#

Backups are not part of GKE pricing directly. Backup for GKE is a separate Google service with a backup management fee (based on the number of non-system namespaces protected) and a backup storage fee (per GiB stored), per the GKE pricing page. Its volume backups are Persistent Disk snapshots, so the snapshot charges above apply to that data.

Logging and Monitoring#

GKE ships with built-in monitoring and logging through Cloud Operations. Cloud Logging includes 50 GiB per project per month. Ingestion beyond that is billed per GiB ($0.50/GiB at the time of writing; see Cloud Logging pricing).

Reading Your Bill#

Monthly totals hide a lot. Use Kubernetes cost monitoring to track per-pod or per-team costs so you can catch idle nodes, oversized requests, and underutilized volumes early. They all show up on the bill without matching use.

Many teams also use Kubernetes cost optimization tools that give granular visibility and tune resources based on real usage. For a side-by-side look at the main platforms, see our guide to Kubernetes cost optimization tools.

Discounts You Can Use#

Google Cloud has a few ways to cut costs if your usage is predictable or can tolerate interruptions.

Sustained Use Discounts (SUDs)#

  • SUDs apply automatically to eligible Compute Engine VMs, including the VMs behind GKE Standard node pools.
  • They start after a resource is used for more than 25% of a billing month.
  • The maximum discount is 30% for N1 (and M1/M2) machine types and 20% for N2, N2D, and C2. E2 machine types are not eligible (Google's sustained use discount documentation, fetched October 8, 2026).
  • Usage already covered by a committed use discount does not also receive a sustained use discount.

Committed Use Discounts (CUDs)#

  • You reserve specific VM types and usage levels for 1 or 3 years.
  • The discount is up to 55% vs pay-as-you-go for the general-purpose e2/n2 families in the live tables above (3-year term). Google's published cap of up to 70% applies only to select memory-optimized machine types not covered here, per Google's CUD documentation.
  • CUDs are available for most node types, including GPU VMs.
  • They fit production workloads with steady usage.

Spot VMs#

  • Spot VMs are temporary and interruptible. Google says Spot discounts go up to 91% for many machine types (Spot VMs). The Spot table above shows about 40% for these e2 and n2 SKUs at the October 8, 2026 snapshot.
  • They work well for stateless or batch workloads.
  • They're available in Standard mode (Spot VMs as nodes) and in Autopilot mode (Spot Pods, billed at the Autopilot Spot rate on Google's GKE pricing page, for example $0.0133 per vCPU per hour in us-central1).

Splitting Workloads Across Node Pools#

You can isolate workload types into separate node pools. Put steady workloads on CUD-covered capacity, run batch or bursty jobs on Spot VMs, and use labels and taints to control scheduling.

Autopilot And Standard Side By Side#

Here is the billing math for a small workload in each mode. The two examples use different workload sizes on purpose. They show how each model is calculated, and they are not a matched-capacity comparison. For a Standard example that adds a control plane, disk, and load balancer, see the worked example below.

Autopilot Workload#

Run 4 pods, each requesting 500m vCPU and 1 GiB of memory, 24/7 for a month (~730 hours, the average month).

Autopilot charges for requested resources only, with no fixed per-pod overhead. A container that requests no ephemeral storage gets a 1 GiB default, so the example includes it. Per pod:

ResourceRateQuantityTotal
vCPU$0.0445/hr0.5$0.02225/hr
Memory$0.0049225/hr per GiB1$0.0049225/hr
Ephemeral storage (default)$0.0001389/hr per GiB1$0.0001389/hr
Total/hr$0.0273114
Monthly (~730 hr)$19.94

Multiply by 4 pods: $0.0273114 × 730 × 4 = $79.75/month in pod compute ($79.34 without the ephemeral storage line). Add the $0.10/hour ($73.00/month) control plane fee, same as Standard, and you get $152.75/month. If this is the only cluster on the billing account, the free tier credit covers the fee and the total stays at $79.75/month.

Spot Pods and committed-use discounts are both available on Autopilot pod pricing (see the Discounts section above).

Standard Workload#

For Standard, assume a 3-node pool of e2-standard-2 VMs (2 vCPU, 8 GiB RAM), priced from the on-demand rate in the October 8, 2026 snapshot of the live table above. The table shows $0.0670/hr, rounded. The unrounded rate is $0.06701/hr:

ItemCalculationPrice
Per node (on-demand)$0.06701 × 730$48.92/month
Total (3 nodes)$0.06701 × 3 = $0.20103/hr; × 730$146.75/month

Add $73.00/month for the control plane and you get $219.75/month, or $146.75/month if the free tier credit covers it. E2 machine types are not eligible for sustained use discounts.

Those nodes can also run other workloads, and Spot VMs or CUDs would bring the cost down further.

Comparing the Two#

These two examples request very different amounts of capacity. The Autopilot workload requests 2 vCPU / 4 GiB total across its 4 pods, while the Standard pool provisions 6 vCPU / 24 GiB across its 3 nodes. Their totals at the October 8, 2026 snapshot ($79.75/month in pod compute, or $152.75 with the control plane fee, for Autopilot; $146.75/month in node compute, or $219.75 with the control plane fee, for Standard) aren’t a fair head-to-head because of that mismatch. Size each mode to the same workload before comparing.

As a rule of thumb, Autopilot tends to cost less on small, spiky, or idle-heavy workloads where you'd otherwise pay for unused node capacity. Standard tends to cost less once you're packing nodes tightly or using Spot/CUD discounts, since its per-vCPU and per-GiB rates are lower than Autopilot’s at matched utilization.

A 3-Node Standard Pool At Live Rates#

Here's a small production node pool priced from the live tables above instead of a static list price. Rates are the unrounded us-central1 values in the October 8, 2026 snapshot, and Google list prices were fetched the same day. The setup:

  • 1 GKE Standard cluster (control plane)
  • 3 × n2-standard-4 nodes (4 vCPU, 16 GiB), on-demand, running ~730 hours/month (average month)
  • 200 GiB of SSD Persistent Disk across PVCs and node boot disks
  • 1 Google Cloud Load Balancer for ingress (one forwarding rule)
ComponentBasisMonthly Cost
Control plane$0.10/hr × 730$73.00
GCE nodes3 × $0.19424/hr (n2-standard-4 on-demand) × 730$425.39
SSD Persistent Disk200 GiB × $0.17/GiB-month (Persistent Disk pricing)$34.00
Load balancer1 forwarding rule, $0.025/hr (first 5 rules) × 730 (Cloud Load Balancing pricing); data processing not included$18.25
Total$550.64

A zonal cluster’s control plane fee is covered by the GKE free tier credit ($74.40/month, one cluster per billing account), so in that case the total drops by $73.00 to $477.64/month. Neither total includes load balancer data processing, egress, snapshots, or logging. The node line is fixed to the snapshot, so the live On-Demand table above may show a different rate.

If you swap the three on-demand nodes for Spot using the Spot table, the GCE line drops by roughly the Spot discount shown there. Move the baseline to a 3-year CUD and it drops by the discount in the Committed Use Discount table instead. The control plane, disk, and load balancer lines don't change. In this example they add up to $125.25 of the $550.64.

Which Mode To Pick#

The choice mostly comes down to how much of the infrastructure you want to own. Do you prefer simplicity or control? Is your usage predictable or bursty? Is this dev/test or production?

When To Pick Autopilot#

Pick Autopilot if:

  • You don't want to manage nodes, scaling, or infrastructure.
  • You want billing tied directly to pod requests.
  • Your workloads are short-lived, dev/test, or steady-state.
  • You're a small team that cares more about shipping than tuning nodes.
  • You're fine with pod-based billing instead of fine-grained node control.

It tends to work well for internal tools, backend services, lightweight event-driven apps, and teams without a dedicated platform engineer.

When To Pick Standard#

Pick Standard if:

  • You want full control over VM types, node pools, and scheduling.
  • You need Spot, custom machine types, or GPUs.
  • You run bursty, compute-heavy, or ML/AI workloads.
  • You have large production deployments and want to tune cost.
  • You use multi-cluster strategies or node isolation patterns.

It suits cost-sensitive production teams, high-performance or GPU workloads, and teams building custom infrastructure on top of Kubernetes.

Running Both#

Some teams run both: Autopilot for dev/test and staging clusters, and Standard for production, GPU, or Spot-heavy workloads.

To go from reading the bill to cutting it, see our GKE cost optimization guide. For the AWS and Azure equivalents of this breakdown, see the EKS pricing guide and the AKS pricing guide.

Frequently Asked Questions#

How much does a GKE cluster cost per month?#

Every cluster pays $0.10 per hour for the control plane, about $73.00 a month. On top of that you pay for compute: per pod request in Autopilot, per VM in Standard. A zonal cluster's control plane fee is covered by the free tier credit, so it can come out to $0 for that line.

Is GKE Autopilot cheaper than Standard?#

It depends on how full your nodes are. Autopilot bills only for what your pods request, so it wins when Standard nodes would sit half empty. Standard wins when you can pack nodes tightly, use Spot, or buy committed use discounts. The side-by-side example in this guide shows both totals.

Do I pay for the GKE control plane?#

Yes, $0.10 per hour per cluster in both modes. The free tier credit of $74.40 a month covers one zonal or Autopilot cluster per billing account, which is why many small setups see no control plane charge.

Are Spot VMs available on GKE?#

Yes, on Standard node pools and as Spot Pods in Autopilot. Google can reclaim Spot capacity at any time with little or no advance notice, so use them for workloads that tolerate interruption.

What costs extra on GKE besides compute?#

Persistent disks, snapshots and backups, network egress, load balancer forwarding rules and their data processing, and logging and monitoring ingestion. These are billed separately in both modes.

Share:
DT

DevZero Team

Cut Kubernetes Cost Before You Pay a Cent.

Every feature unlocked. No hidden fees.

Start for free

Start Free

$0/ month
Unlimited clusters
K8s resource & cost monitoring
Network monitoring
Cost attribution for departments
Multi-cloud support & governance
Audit logging