Releases

Kubeflow 26.03.1 is a large calendar release bundling upgrades across Pipelines, Kserve, Trainer, Notebooks, Dashboard, Istio, Knative, Dex, cert-manager, and oauth2-proxy, alongside a breaking rename of model-registry to hub and a relocation of model registry/catalog namespaces that operators must account for before upgrading.

• breakingmodel-registry package renamed to hub

  The kubeflow/model-registry package has been renamed to kubeflow/hub. Update manifests, kustomizations, and any automation referencing the old package name.

• breakingModel registry now deploys to user profile namespace

  Model registry deployments now land in the user's profile namespace instead of the shared default namespace. Review per-user access, RBAC, and any scripts that assumed a fixed default-namespace location.

• breakingModel catalog moved to cluster-wide singleton

  The model catalog is now split out into the kubeflow namespace as a single cluster-wide instance rather than one per namespace. Check for per-namespace catalog customizations that would need consolidation.

• enhancementModel registry with UI now on by default

  Model registry with its UI is now enabled by default. Clusters that previously left it off will get it running after upgrade unless explicitly disabled.

KServe v0.19.0 is a large release focused heavily on LLMInferenceService (LLMISvc) maturity: better observability, autoscaling, routing, and a migration path for llm-d v0.6 upgrades. Two CVEs are patched.

• securityPatch two CVEs before upgrading or apply independently

  v0.19.0 patches a vLLM/Pillow CVE and pins azure-core>=1.38.0 for CVE-2026-21226. If you are running KServe with vLLM runtimes or azure-core dependencies, upgrade to v0.19.0 or manually apply the azure-core pin to your environment. Check your current Pillow version in custom server images as well.

• breakingRouter splitter off-by-one fix changes traffic split behavior

  The pickupRoute random range had an off-by-one error. After upgrading, traffic percentages in InferenceService splitter configurations will be computed correctly, which means observed traffic distribution will shift if your weights were tuned around the buggy behavior. Verify split configurations in staging before rolling to production.

• enhancementLLMISvc autoscaling and status visibility are now production-ready

  HPA/KEDA scaling conditions are now visible in LLMISvc status, and readiness transitions emit k8s events. If you are operating LLMInferenceService, add these conditions to your monitoring dashboards and alerting. The --scaling flag in kserve-install.sh simplifies enabling autoscaling during fresh installs.