GitHub Copilot retires six models Oct 19: GPT-5.5 and Grok 4.5
The cuts hit Chat, agent, and completions at once — as Copilot adds cost-tier auto-select and a Sentry crash-to-PR canvas, and json-render ships routed apps.
Copy markdown
Six models, gone in four weeks
On Oct 19 Copilot drops GPT-5.5, GPT-5.4 and 5.4 mini, GPT-5 mini, Gemini 3.7 Flash, and Grok 4.5 from Chat, agent mode, and code completions. They map to GPT-5.6 Sol/Luna, Gemini 3.8 Flash, and Grok 4.6 — so anything pinned to an old model name stops resolving that day.
Enterprise auto-swaps; solo plans don't
Business and Enterprise seats switch to the replacement models automatically unless an admin turned off default enablement, but Pro and Free users must enable the new model themselves. Check your model picker before the 19th so an agent run doesn't silently lose its model.
Auto-select now trades cost against quality
The same weekly release added efficiency, balance, and intelligence tiers to Copilot's automatic model selection, letting you bias the router toward cheaper or smarter without hand-picking a model per request. Useful for capping spend on high-volume completions.
Elsewhere: crash report to pull request, one canvas
A new Sentry canvas in the Copilot app takes you from a production error to a prepared PR, and VS Code agents can now run in local Dev Containers and auto-close once their PR merges. Fewer hops between your error tracker and your editor.
Elsewhere: one JSON spec, a whole routed app
Vercel's json-render 0.21.0 adds a TanStack Start renderer that turns a guardrailed JSON spec into a full app — file-based routes, SSR loaders, head metadata, and client navigation — from a component catalog you define. A weekend route to LLM-driven UIs that can't render components you didn't approve.