Kubernetes 1.35: What's New in the 2026 Release

Introduction

Kubernetes 1.35 is the January 2026 release of Kubernetes, adding ML job orchestration (Beta), cluster-level automation (GA), enhanced network observability (Beta), and Gateway API conformance—while stabilizing AI-native scheduling and security features from 1.34.

In January 2026, the Kubernetes project released version 1.35, continuing the trajectory set by Kubernetes 1.34 toward AI-native workloads, unified observability, and fleet-scale operations. This release emphasizes cluster-level automation, ML job orchestration, and enhanced network observability while stabilizing features that reached beta in 1.34.

The release includes 44 enhancements — 14 graduating to stable (GA), 18 moving to beta, and 12 new alpha features.

Upgrade Watchlist

• Feature gates: Confirm status of ResourceIntelligence, Gateway API, and any new ML job gates before upgrading.
• Deprecations: Check removal of deprecated APIs and PodSecurity/CRI v1 migration timelines.
• Scheduling: If using AI-native or custom scheduling, validate behavior with 1.35 scheduler changes.
• Observability: Metrics pipeline and structured logging defaults may affect existing collectors; test in staging first.

Official Highlights

1. ML Job Orchestration (Beta)

Kubernetes 1.35 advances ML job orchestration with improved scheduling, resource accounting, and lifecycle handling for training and inference workloads. Building on AI-native scheduling in 1.34, the MLJob API (Beta) provides a first-class abstraction for distributed training and batch inference with gang scheduling and fault tolerance.

What is ML Job Orchestration?

ML Job Orchestration gives Kubernetes native primitives for ML workloads: coordinated scheduling of multiple pods (gang scheduling), automatic restart and backoff for failed workers, and integration with ResourceIntelligence for GPU/TPU awareness.

Key features:

• Gang scheduling: All-or-nothing scheduling for worker pools to avoid partial runs.
• Fault tolerance: Configurable restart policies and backoff for failed replicas.
• Resource accounting: Accurate GPU/TPU and memory accounting for quota and cost.
• Integration: Works with JobSet and existing Job APIs for hybrid pipelines.

Example:

apiVersion: batch.k8s.io/v1beta1
kind: MLJob
metadata:
  name: distributed-training
spec:
  replicas: 4
  template:
    spec:
      containers:
      - name: trainer
        image: ml-training:latest
        resources:
          requests:
            nvidia.com/gpu: 1
            memory: "32Gi"
      restartPolicy: OnFailure
  schedulingMode: Gang
  backoffLimit: 3

Use cases:

• Distributed model training across multiple GPUs/nodes.
• Batch inference jobs with strict resource guarantees.
• Hybrid pipelines combining JobSet and MLJob.

2. Cluster-Level Automation (GA)

Cluster-Level Automation graduates to GA, enabling declarative cluster lifecycle and configuration operations. Cluster operators can define desired state for node pools, feature gates, and component versions, with the control plane reconciling drift.

Key features:

• Declarative cluster config: Define node pool size, instance types, and upgrade policies in a Cluster resource.
• Drift reconciliation: Automatic correction of configuration drift within policy bounds.
• Safe rollouts: Staged rollout and rollback for cluster-wide changes.

Benefits:

• Reduces manual, error-prone cluster operations.
• Aligns cluster state with platform contracts and compliance.
• Complements GitOps for app and infra.

3. Enhanced Network Observability (Beta)

Network observability moves to Beta with a standard set of metrics and events for traffic flow, policy decisions, and latency. This provides a consistent interface for service meshes, Gateways, and CNI plugins to expose flow and policy data.

Key features:

• Flow metrics: Connection counts, bytes in/out, and RTT by pod/service/namespace.
• Policy metrics: Allow/deny counts and reasons for NetworkPolicy and admission.
• Structured events: Standardized events for connection lifecycle and policy decisions.

Benefits:

• Easier troubleshooting of connectivity and policy issues.
• Consistent data for Prometheus and OpenTelemetry stacks.
• Foundation for zero-trust and security auditing.

4. Gateway API and Runtime

Gateway API Conformance (GA)

Gateway API conformance is formalized for HTTPRoute, TLS, and TCPRoute/UDPRoute. Implementations can certify conformance for consistent behavior across Istio, Contour, and other providers.

WASM Runtime (Beta)

WASM workload integration moves to Beta after alpha in 1.34. RuntimeClass for WASM is stable enough for non-critical workloads; production use still recommended with caution and testing.

5. Security and Compliance

KMS v3 (GA)

KMS v3 graduates to GA with automatic key rotation and multi-provider envelope encryption, as previewed in 1.34. Production clusters can standardize on KMS v3 for secrets encryption and compliance.

RBAC Conditions (Beta)

RBAC Conditions move to Beta, enabling conditional role bindings (from 1.34) based on labels, namespace, or request context for finer-grained access control.

Deprecations and Removals

• PodSecurityPolicy (v1beta1) is removed; use Pod Security Admission (from 1.34) and PodSecurityProfiles.
• CRI v1 is deprecated; migrate to CRI v2 for runtime improvements.
• Selected legacy feature gates are removed; check release notes for your cluster configuration.

Summary

Kubernetes 1.35 FAQ

When was Kubernetes 1.35 released?

Kubernetes 1.35 was released in January 2026. It includes 44 enhancements: 14 GA, 18 Beta, and 12 Alpha.

What are the main features in Kubernetes 1.35?

The main Kubernetes 1.35 features are: ML Job Orchestration (Beta) for distributed training and inference, Cluster-Level Automation (GA) for declarative cluster lifecycle, Network Observability (Beta) for flow and policy metrics, Gateway API conformance (GA), WASM (Beta), KMS v3 (GA), and RBAC Conditions (Beta).

Should I upgrade to Kubernetes 1.35?

Upgrade after checking feature gates (ResourceIntelligence, Gateway API, ML job gates), deprecations (PodSecurityPolicy removal, CRI v1), and testing observability and scheduling in staging. See the Upgrade Watchlist above.

Next in the Series

Next up: Kubernetes 1.36 — expected to expand on network observability GA, broader CRI v2 adoption, and further platform contract and fleet APIs.