KubeVirt vs GitHub

Two moves dominate the current window and they point in opposite directions. GitHub Models is fully retired - playground, model catalog, inference API and bring-your-own-key all gone for every customer - while stacked pull requests entered public preview, changing the unit of code review from a single large diff to an ordered series of layers. Around them sits the familiar Copilot governance work: user-based model policy targeting for Enterprise, default model enablement so admins stop toggling each new model, and device restrictions on remotely controlled Copilot sessions. The curated model shelf keeps turning over too, with Gemini 2.5 Pro and Gemini 3 Flash pulled from every Copilot surface. Security hardening continues in parallel, with npm granular access tokens losing their 2FA bypass for sensitive operations.

GitHub is narrowing where it competes. Retiring Models concedes raw model access and inference to the model vendors, while investment concentrates on the surfaces GitHub actually owns - the repository, the review flow, and the administrative controls enterprises need before they will turn any of it on. The Copilot line has settled into a reach-then-govern rhythm: ship capability broadly, then follow with policy targeting, defaults and device scoping. Model deprecations landing on the same day as policy controls show the two halves of that position - GitHub curates the shelf and governs access to it, rather than competing on the models themselves. Stacked pull requests are the clearest sign that core platform workflow, not AI surface area, is where the differentiation is being placed.

Expect stacked pull requests to pick up API and CLI support before general availability, and the model-governance controls to consolidate into a single policy surface now that per-model manual enablement has been replaced by defaults.