Six GitHub Copilot models were deprecated — replacements still need policy and task checks

GitHub’s [October 19 notice](https://github.blog/changelog/2026‑10‑19‑github‑copilot‑models‑deprecated‑on‑october‑19‑2026/) says Gemini 3.7 Flash, GPT‑5.5, GPT‑5.4, GPT‑5.4 mini, GPT‑5 mini and Grok 4.5 are deprecated across most Copilot experiences. GitHub explicitly names chat, inline edits, ask and agent modes, and code completions among the affected surfaces. GitHub had announced the change on September 18, so this is the effective retirement date rather than a new future warning. Users do not need to remove the old models manually, but workflows that selected them should no longer assume the choice remains available. The first inventory step is practical: search shared Copilot instructions, team documentation and automation for pinned model names, then ask frequent users which selector choices they rely on. A model can disappear from the product even when no repository file mentions it.

GitHub points Gemini 3.7 Flash users to Gemini 3.8 Flash; GPT‑5.5 and GPT‑5.4 users to GPT‑6.1 Sol; GPT‑5.4 mini and GPT‑5 mini users to GPT‑6 Luna; and Grok 4.5 users to Grok 4.7. Those suggestions establish the supported path inside Copilot. They do not establish identical latency, context handling, tool behavior, cost accounting or output quality for every codebase. A newer model can be stronger overall and still change a fragile workflow. Treat the table as the start of acceptance. Re‑run a small set of representative tasks: explain a repository‑specific pattern, propose a bounded edit, follow local instructions, execute the relevant tests and avoid touching unrelated files. Retain outputs and rework rather than relying only on subjective preference.

GitHub warns that Copilot Enterprise administrators may need to enable a replacement through model policies. The company recommends checking individual Copilot settings and confirming that the policy for the specific model is enabled; an enabled model should then appear in the selector in VS Code and on GitHub.com. That creates two separate failure modes. A user may be mapped to a valid replacement that the organization has not allowed, or the policy may be enabled while a local client has not refreshed its available list. Record both policy state and client acceptance. The new model‑policy API can help teams read and update those controls, but a migration should still use expected‑state writes and a narrow rollout. Do not enable every model everywhere simply because one replacement is needed.

A durable Copilot workflow defines its job through repository instructions, allowed tools, review rules and tests. The model is one implementation choice inside that contract. When a model retires, the team should be able to substitute a supported option and demonstrate that the contract still holds. If a replacement fails an important task, narrow its use, revise the workflow or choose another supported model. Do not hide the failure by lowering tests or skipping human review on agent changes. Record the specific mismatch so a later client or model update has a finite retest condition. The retirement is already effective, so affected teams should prioritize access and task validation. The release is not evidence that every user needs the same replacement or that GitHub’s suggested pairings preserve all behavior. *Figure by OpenTools Team; no third‑party expressive material used. It separates GitHub’s replacement mapping from policy enablement and task‑level acceptance.*