# Criteria for Selecting GKE Autopilot vs Standard Mode: Complete Decision Guide > Decide between GKE Autopilot and Standard mode. Learn the criteria for selecting the right GKE mode based on your workload needs for node management, billing, and control. - Repository: [Google/skills](https://github.com/google/skills) - Tags: deep-dive - Published: 2026-08-09 --- **Choose GKE Autopilot for fully managed node operations and per-pod billing when you do not require custom kernel parameters or privileged DaemonSets; select Standard mode when you need complete control over node pools, custom OS images, or specific kernel-level configurations.** According to the `google/skills` repository, selecting the appropriate cluster mode requires evaluating architectural trade-offs between operational overhead and infrastructure flexibility. The criteria for selecting GKE Autopilot vs Standard mode depend on whether your workloads align with Google's golden-path defaults or demand low-level customization of the underlying compute layer. ## Core Selection Criteria The primary distinctions between these modes are defined in [`skills/cloud/gke-basics/references/core-concepts.md`](https://github.com/google/skills/blob/main/skills/cloud/gke-basics/references/core-concepts.md) and detailed in the mode selection guidance within [`skills/cloud/gke-cluster-creation/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gke-cluster-creation/SKILL.md). **Node Management** - **Autopilot**: Google fully manages the underlying VMs; you never interact with node pools or machine types directly. - **Standard**: You maintain complete responsibility for node pool configuration, machine types, operating system images, and autoscaling policies. **Pricing Structure** - **Autopilot**: Pay per pod resource request (CPU, memory, ephemeral storage) with no separate cluster management fee. - **Standard**: Pay for underlying Compute Engine VMs plus a per-cluster management fee regardless of utilization. **Customization Capabilities** - **Autopilot**: Customization occurs through **ComputeClasses**, allowing selection of predefined machine families, Spot VMs, and GPU targeting without node-level access. - **Standard**: Full control over kernel parameters, DaemonSets, custom OS images, sysctls, and privileged containers. **Default Recommendation** - **Autopilot**: Serves as the default for new clusters and represents the recommended golden path for most production use cases. - **Standard**: Reserved for scenarios where documented Autopilot limitations apply. ## Architectural Trade-offs When evaluating the criteria for selecting GKE Autopilot vs Standard mode, consider these five architectural dimensions: **Operational Overhead** Autopilot eliminates node lifecycle management—Google automatically patches, upgrades, and scales nodes based on pod requests. Standard mode requires manual maintenance of node health, OS patches, and autoscaling configuration. **Cost Efficiency** Autopilot charges only for requested pod resources, making it economical for bursty or stateless workloads with variable utilization. Standard mode requires paying for full VM capacity, which may reduce costs for consistently high-utilization workloads that maximize node capacity. **Security Posture** Autopilot enforces restrictive security defaults including Shielded Nodes and Pod Security Standards out-of-the-box. Standard mode provides flexibility to run privileged DaemonSets or custom kernel modules when necessary, though this requires manual security configuration. **Hardware Access** GPU and TPU workloads are supported in Autopilot via ComputeClasses, but Standard mode offers granular control over specific driver versions, custom images, and specialized networking plugins not yet available in Autopilot. **Availability Topology** Autopilot clusters are regional by default, providing high-availability control planes across three zones. Standard clusters support both regional and zonal configurations, allowing cost-optimized single-zone control planes for development and testing environments. ## When to Choose GKE Autopilot Select Autopilot when your workloads match these criteria: - You run production workloads that do not require custom kernel tuning or privileged system-level DaemonSets. - Your team wants to minimize operational overhead and focus exclusively on application logic rather than infrastructure management. - You operate bursty, stateless, or cost-sensitive workloads where pod-level billing provides better resource efficiency than VM-based pricing. ## When to Choose GKE Standard Standard mode is the appropriate choice when: - Your workloads require custom node OS configurations, specific kernel parameters, or privileged system-level components. - You must control exact VM types, attach custom GPUs with specific driver requirements, or run specialized networking plugins unsupported in Autopilot. - You run long-running, high-utilization workloads where full node utilization reduces per-resource costs compared to per-pod billing. ## Implementation Examples The following commands demonstrate the creation of both cluster types as documented in the google/skills repository. Create an Autopilot cluster with comprehensive monitoring and security features: ```bash gcloud container clusters create-auto my-autopilot-cluster \ --region us-central1 \ --project $PROJECT_ID \ --release-channel regular \ --enable-private-nodes \ --enable-master-authorized-networks \ --enable-dns-access \ --enable-secret-manager \ --scoped-rbs-bindings \ --monitoring=SYSTEM,API_SERVER,SCHEDULER,CONTROLLER_MANAGER,STORAGE,POD,DEPLOYMENT,STATEFULSET,DAEMONSET,HPA,CADVISOR,KUBELET,DCGM \ --quiet ``` Create a Standard regional cluster with autoscaling and shielded nodes: ```bash gcloud container clusters create my-standard-cluster \ --region us-central1 \ --project $PROJECT_ID \ --num-nodes 3 \ --machine-type e2-standard-4 \ --disk-type pd-balanced \ --enable-autoscaling --min-nodes 1 --max-nodes 10 \ --enable-shielded-nodes --enable-secure-boot \ --workload-pool=$PROJECT_ID.svc.id.goog \ --enable-private-nodes \ --enable-master-authorized-networks ``` ## Summary - **Autopilot** provides fully managed nodes with per-pod billing and is the default recommendation for most production workloads in the `google/skills` reference architecture. - **Standard** mode offers complete node-level control but requires manual management of node pools, OS patching, and autoscaling policies. - Choose Autopilot when you prioritize operational simplicity and have workloads compatible with ComputeClasses and default security constraints. - Choose Standard when you require kernel-level customization, custom OS images, or specialized hardware configurations not supported by Autopilot. - Reference the selection criteria in [`skills/cloud/gke-cluster-creation/SKILL.md`](https://github.com/google/skills/blob/main/skills/cloud/gke-cluster-creation/SKILL.md) and the core concepts in [`skills/cloud/gke-basics/references/core-concepts.md`](https://github.com/google/skills/blob/main/skills/cloud/gke-basics/references/core-concepts.md) for detailed implementation guidance. ## Frequently Asked Questions ### What is the main pricing difference between GKE Autopilot and Standard? Autopilot charges based on pod resource requests (CPU, memory, ephemeral storage) with no cluster management fee, while Standard requires payment for the underlying Compute Engine VMs plus a per-cluster management fee regardless of actual pod utilization. ### Can I use GPUs with GKE Autopilot? Yes, Autopilot supports GPU and TPU workloads through **ComputeClasses**, which allow targeting specific machine families and accelerator types; however, Standard mode provides greater flexibility for custom driver versions and specialized GPU configurations. ### Why is Autopilot considered the "golden path" for GKE? According to the source code in [`skills/cloud/gke-basics/references/core-concepts.md`](https://github.com/google/skills/blob/main/skills/cloud/gke-basics/references/core-concepts.md), Autopilot is the default for new clusters because it eliminates node management overhead, enforces secure defaults like Shielded Nodes and Pod Security Standards, and optimizes costs for typical production workloads. ### When must I use Standard mode instead of Autopilot? You must use Standard mode when your workloads require custom kernel parameters, privileged DaemonSets, custom OS images, specific sysctl configurations, or specialized networking plugins that are not supported by Autopilot's managed node environment.