October 19, 2026 is the date GitHub set for retiring six models from Copilot. Its September 18 announcement covers Chat, inline edits, ask and agent modes, and code completions. Treat this as a scheduled platform change, not proof that every client has already completed the transition.
Which replacements did GitHub recommend?
| Scheduled for retirement | Suggested alternative |
|---|---|
| Gemini 3.7 Flash | Gemini 3.8 Flash |
| GPT-5.5 and GPT-5.4 | GPT-5.6 Sol |
| GPT-5.4 mini and GPT-5 mini | GPT-5.6 Luna |
| Grok 4.5 | Grok 4.6 |
These are GitHub's suggested alternatives, not a claim that each replacement produces equivalent results. The notice concerns Copilot availability; it is not evidence of a shutdown of those providers' direct APIs. Source: GitHub retirement notice.
Why the model picker is only one check
GitHub's access documentation says availability depends on the subscription, client and organization or enterprise restrictions. Owners can enable or disable models for Business and Enterprise seats. A replacement being offered by the platform therefore does not establish that a particular team can use it.
The default-availability documentation describes an already active model policy: unconfigured eligible models follow the default unless explicitly configured. Do not confuse that with the separate feature policy scheduled to activate on October 22.
The documentation is not fully aligned
When checked for this article, GitHub's supported-models reference still listed the affected models, and its retirement-history table did not include this six-model wave. The dated changelog provides the announced deadline; the reference page does not independently confirm completed removal. Neither source establishes the live state of your organization's clients.
The same reference explains that availability differs by plan and client, and that some models require minimum IDE or extension versions. Check the actual environment rather than treating a documentation table as a live status feed.
A practical validation approach
Our recommendation is to separate access validation from output validation. First, ask an administrator to confirm the intended replacement is permitted. Then open the client developers actually use and verify that it is selectable. Record the client version and policy decision so troubleshooting has a concrete starting point.
Next, rerun a small set of representative tasks: a routine edit, a multi-file change and a test-writing request. Compare correctness, review effort and any usage information available to your team. These are suggested checks, not GitHub requirements or a prediction of degraded performance.
Keep the change narrowly scoped. A Copilot retirement notice should trigger a Copilot review. For separate provider integrations, consult that provider's own lifecycle notices before changing production dependencies.
The operational lesson is straightforward: assign ownership for model changes, keep a repeatable acceptance test and distinguish platform availability from organization approval. A migration is finished when the workflow is verified, not merely when a replacement name is chosen.