Skip to main content

Anyscale platform release notes - March 2026

View Markdown

Anyscale platform release notes - March 2026

Use this page to learn about new features and improvements to the Anyscale platform released in March 2026, including the Anyscale console, CLI, and SDK.

To learn about new features added as part of Ray releases, see Ray release highlights.

To learn about Python and system libraries installed in each Anyscale base image, see Anyscale base images.

To view platform release notes from other months, see All platform release notes.

Anyscale operator for Kubernetes version 1.5.0 released​

March 18, 2026

Anyscale has released version 1.5.0 of the Anyscale operator for Kubernetes. See the Anyscale operator GitHub page for release details.

  • Gateway API support: The operator now supports the Kubernetes Gateway API with envoy and istio controllers. See Networking configuration for configuration details.
  • Kueue deadlock fix: Fixed an issue where workload termination while a pod is in SchedulingGated state caused a deadlock during pod and workload cleanup.

Anyscale CLI version 0.26.91 released​

March 17, 2026

A new version of the Anyscale CLI and SDK is now available. See Version 0.26.91 release notes.

Updated default head node size for Kubernetes deployments​

March 17, 2026

For Anyscale clouds deployed on Kubernetes, the Anyscale console now defaults the head node to 8CPU-32GB, matching the default for VM-based clouds. The console also displays a warning when you select a head node with fewer than 8 CPUs or less than 32 GiB of memory.

For head node sizing guidance, see Node sizing guidelines.

Anyscale operator for Kubernetes version 1.4.2 released​

March 12, 2026

Anyscale has released version 1.4.2 of the Anyscale operator for Kubernetes. For release details, see the Anyscale operator GitHub page.

Changes in 1.4.2:

  • Cross-namespace permission checks: The operator now runs Kubernetes permission checks across all namespaces defined in the workloads.managedNamespaces configuration. Status details show which namespaces the operator successfully verified. View these details in the Anyscale console or with anyscale cloud status.

Anyscale CLI version 0.26.90 released​

March 9, 2026

A new version of the Anyscale CLI and SDK is now available. See Version 0.26.90 release notes.

Update Anyscale clouds backed by Google Cloud VMs​

March 9, 2026

Anyscale clouds deployed on Google Cloud VMs using anyscale cloud setup or anyscale cloud resource setup might have a dependency on Google Deployment Manager, which will retire on June 30, 2026. Update your cloud today to ensure Anyscale can clean up resources in your Google Cloud account. See Update Anyscale clouds deployed with Deployment Manager.

Anyscale CLI version 0.26.89 released​

March 4, 2026

A new version of the Anyscale CLI and SDK is now available. See Version 0.26.89 release notes.

Deprecation of legacy GPU instances​

March 2, 2026

Anyscale has adopted Linux open GPU kernel modules for VM images, which don't support the following accelerator types: K80, M60, V100 16GB, and V100 32GB. Instances types using these GPUs are now deprecated for Anyscale clouds on the VM stack. For AWS, the g3s, g3, p3, and p3dn families are deprecated. For Google Cloud, instance types in the n1 family that use nvidia-k80 or nvidia-v100 are deprecated.

Anyscale CLI version 0.26.88 released​

March 2, 2026

A new version of the Anyscale CLI and SDK is now available. See Version 0.26.88 release notes.

Removal of deprecated SDK methods​

March 2, 2026

Anyscale CLI version 0.26.88 removes support for legacy job, schedule, cluster, and image SDK methods on AnyscaleSDK. Use the modern anyscale.job.*, anyscale.schedule.*, workspace, and API client alternatives. See Methods removed in version 0.26.88 for migration guidance.

Declarative compute configs for VM-based clouds (beta)​

March 2, 2026

You can now use declarative compute configs for Anyscale clouds backed by AWS EC2 or GCP GCE. Specify CPU, memory, and GPU requirements with required_resources and optional required_labels for head and worker nodes; Anyscale selects and provisions VM instance types that match. This feature is in beta release. See Declarative compute configs.