--- title: "What's new in Lens K8S IDE 2026.9: Manage GPU and NIC devices with DRA — Lens Blog — Lens" description: "Lens K8S IDE 2026.9 brings Kubernetes DRA to the navigator for managing devices like GPUs and NICs, plus an advanced editor for editing YAML manifests." url: https://lenshq.io/blog/lens-release-september26/ --- [← Blog](https://lenshq.io/blog/) Lens Release # What's new in Lens K8S IDE 2026.9: Manage GPU and NIC devices with DRA Yalton Ruiz September 2, 2026 4 min read **TL;DR:** - **Device Resources** lands in the navigator: Kubernetes DRA support with claim, template, device class, and slice visibility, per-pod and per-node. - **Advanced editor** adds a side-by-side diff view and one-click **Revert to original** when editing a live resource. - **Ask AI now has a tool selector** in the top bar: switch between Claude Code, GitHub Copilot CLI, Gemini CLI, Codex CLI, OpenCode, or a custom command, right when you need to. Kubernetes Dynamic Resource Allocation (DRA) reaches into the navigator this release as **Device Resources**, a diff-mode editor makes live edits reversible, and the Ask AI tool selector lets you pick your agent on the fly. **Kubernetes Dynamic Resource Allocation reached general availability in Kubernetes 1.34**, according to the [official Kubernetes blog](https://kubernetes.io/blog/2025/09/01/kubernetes-v1-34-dra-updates/). ## Device Resources for Kubernetes DRA DRA is a new way for a workload to request specialized hardware, a GPU, a TPU, an FPGA, a high-speed NIC, by its properties (memory size, driver, topology) instead of just asking for "1 GPU" and hoping for the best. GPU-heavy AI and inference workloads are the most common reason teams need this today, but DRA isn't limited to GPUs: it covers any specialized device a cluster hands out through a DRA driver, including Google's own TPU driver. In short: hardware gets listed, sorted into classes, and matched to whatever asks for it. ![Four-step diagram: a driver reports what hardware exists (ResourceSlice), a rule sorts it into device classes (DeviceClass), a pod's device claim names the class it needs (ResourceClaim), and Kubernetes matches it to a real device (Pod)](https://lenshq.io/blog/lens-release-september26/dra-object-flow-diagram.webp) Lens K8S IDE 2026.9 brings a **Device Resources** section into the navigator, built on Kubernetes' ResourceClaim and ResourceSlice APIs: - Browse Resource Claims, Resource Claim Templates, Device Classes, and Resource Slices, with detail panels for each. - Open a pod and see exactly which claims, and which devices, it's holding. - Open a node and see its full device inventory from the Resource Slices it publishes. - Get warned the moment a device reports itself unhealthy, instead of finding out from a failed workload. - Ask AI about any DRA resource, instead of writing `kubectl` one-liners to cross-reference claims and slices. - Create Resource Claims, Templates, and Device Classes from built-in templates. ![The Device Resources navigator group expanded, showing Resource Claims, Resource Claim Templates, Device Classes, and Resource Slices, with a Resource Claims list open for a cluster with GPU claims](https://lenshq.io/blog/lens-release-september26/device-resources-claims-view.webp) Device Resources is available now to everyone running Lens K8S IDE, no premium tier required. ## A safer way to edit: the advanced editor Editing a live Kubernetes resource means changing cluster state directly, with no undo. The advanced editor gives that edit a safety net: switch to diff mode and see exactly what you've changed since you opened the tab, original on one side, your edits on the other, changed lines highlighted. ![The advanced editor in diff mode while editing a Deployment, with the original resource on the left and the edited version on the right, changed lines highlighted](https://lenshq.io/blog/lens-release-september26/advanced-editor-diff-mode.webp) If something looks wrong, **Revert to original** puts it back in one click. It's filed under Developer Productivity in Premium Features. ## Switch AI tools from the top bar, on the fly If you use an AI coding assistant alongside your cluster work, you probably reach for a different one depending on the task. Ask AI's tool button now opens a selector, so you can pick whichever agent fits the moment instead of being locked into whatever you set in preferences. ![The Ask AI tool button's menu listing Claude Code, GitHub Copilot CLI, Codex CLI, OpenCode, Custom, and Gemini CLI, with a tick on the tool in use](https://lenshq.io/blog/lens-release-september26/ai-tool-selector-menu.webp) Pick Claude Code, GitHub Copilot CLI, Gemini CLI, Codex CLI, OpenCode, or your own custom command directly from the top bar. Each option shows whether it's on your `PATH`, with an install link or a Configure button where one is needed. ## Also in This Release - Cluster connection for non-admin users no longer stalls at "Refreshing cluster accessibility…" for minutes or hours; the check now completes in seconds. - Argo CD resources (Applications, ApplicationSets, AppProjects) now show up in Custom Resources even when the Argo CD namespace isn't part of the tab's namespace selection. - Custom resource lists no longer occasionally stay empty until you reopen the view. - Argo CD and Flux dashboards no longer wrongly claim limited namespace visibility on clusters with full access. ## Get the update Lens K8S IDE 2026.9 is available now, or update from inside the app. The advanced editor and the Ask AI tool selector are Premium Features; see [lenshq.io/pricing](https://lenshq.io/pricing/) for plan details. ## Get Lens K8S IDE 2026.9 [See pricing](https://lenshq.io/pricing/) [Download](https://lenshq.io/download/)