generated: '2026-07-21'
method: searched
source: https://run-ai-docs.nvidia.com/saas/getting-started/whats-new-for-nvidia-run-ai-saas.md
docs:
  saas: https://run-ai-docs.nvidia.com/saas/getting-started/whats-new-for-nvidia-run-ai-saas.md
  self_hosted: https://run-ai-docs.nvidia.com/self-hosted/getting-started/whats-new.md
scheme: dated releases (SaaS ~every two weeks) + product versions (self-hosted 2.x)
current_self_hosted_version: '2.26'
cadence: >-
  SaaS features are gradually rolled out ~every two weeks; all customers receive
  changes within 10 days of initial release. Hotfixes deploy immediately.
entries:
- date: '2026-07-05'
  track: saas
  highlights:
  - Kubernetes 1.36 support (from cluster v2.26 onward).
  - Fix for batch job workload types stuck in Resuming state (requires K8s v1.32.9/1.33.6/1.34.2/1.35.0+).
- date: '2026-06-09'
  track: saas
  highlights:
  - Resolved multiple HIGH/CRITICAL security vulnerabilities (CVE fixes).
  - Fixed CLI accepting unsupported storage flags for distributed inference submissions.
  - Fixed workload Connections dialog showing node IP instead of tool name.
- date: '2026-04-28'
  track: saas
  version: '2.25'
  breaking: false
  highlights:
  - Kubernetes Gateway API support (opt-in alternative to Ingress, zero-downtime migration).
  - Component version updates — OpenShift 4.21, GPU Operator 26.3; removed support for K8s 1.32 / OpenShift 4.17.
- date: '2026-04-15'
  track: saas
  highlights:
  - Additional product enhancements and resolved bugs (see release notes).
note: >-
  Recent window only; the live What's New page is the source of truth. Deprecated
  APIs are announced in these release notes per the API policy.