For supported Kubernetes versions, see the Version Guide. For supported operating systems—including ContainerOS, Alibaba Cloud Linux 4 Container-optimized, Alibaba Cloud Linux 3 Container-optimized, Alibaba Cloud Linux 3, Alibaba Cloud Linux 3 for ARM, Alibaba Cloud Linux UEFI 3, Red Hat, Ubuntu, and Windows—see Operating systems.
February 2026
Kubernetes version support
ACK now supports Kubernetes 1.35.
Starting from Kubernetes 1.35, cgroup v1 is no longer supported. Nodes still using cgroup v1 will fail to start the kubelet. cgroup v2 support became stable in Kubernetes 1.25. Before upgrading to Kubernetes 1.35, upgrade your node operating system to one that supports cgroup v2.
For a full list of changes, see Kubernetes 1.35.
New features
Container Service for Kubernetes
| Feature | Description |
|---|---|
| Agent Mode for ACK AI Assistant | Supports root cause analysis for complex failures, recommends best practices, and simplifies cluster O&M. See Authorize the Agent feature of ACK AI Assistant. |
| ContainerOS v3.7 | Adds a new image with pre-installed GPU drivers. See ContainerOS image release history. |
| LLM service deployment and intelligent routing using Gateway with Inference Extension | Combined with Knative serverless architecture, simplifies managing generative AI inference services. Provides efficient Layer 7 routing and load balancing across multiple inference service workloads, with GPU resource auto scaling based on request concurrency. See Deploy LLM services and implement intelligent routing using Gateway with Inference Extension. |
| Custom admission checks for ACK Kube Queue | By default, ACK Kube Queue checks only the Max quota defined in the ElasticQuotaTree before the scheduler schedules jobs. For more complex dequeuing logic, use the admission check mechanism. See Add custom admission checks for ACK Kube Queue. |
Argo Workflow Cluster
| Feature | Description |
|---|---|
| Argo Workflow Cluster | A fully managed service that optimizes the open-source workflow engine for efficient, stable execution of large-scale workflows. See Argo Workflow Cluster overview. |
January 2026
New features
| Feature | Description |
|---|---|
| CPFS for AI dynamic volumes | Dynamic volume provisioning automates on-demand storage for CPFS for AI, eliminating manual PV management. Volumes support concurrent reads and writes from multiple applications—useful for AI training and big data analytics workloads that share code, configuration files, and intermediate computation results. See Use CPFS for AI dynamic volumes. |
| On-demand pre-filling of OSS data to high-performance volumes | Pre-fill large amounts of cold data from OSS to a high-performance volume (such as a CPFS for AI volume or a cloud disk) before AI training or data analysis tasks begin. Computing tasks then read data directly from the high-performance volume at high speeds. The storage volume is automatically released after the task completes, balancing performance with cost. |
| Hybrid cloud node pools | Register nodes from an on-premises data center to an ACK cluster for centralized management and scheduling of both on-premises and cloud resources. See Schedule applications to a hybrid cloud node pool. |
| yurt-hub | A node component that enables nodes in hybrid cloud node pools to function autonomously in ACK clusters. See yurt-hub. |
Historical release notes
For release notes from 2025 and earlier, see Historical release notes (2025 and earlier).