GitHub Copilot Model Deprecations: What Changed on October 2

GitHub deprecated four Copilot AI models on October 2, 2026 — Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7 — across every Copilot surface, including Copilot Chat, inline edits, ask mode, agent mode, and code completions. The change was announced in advance on GitHub's changelog and confirmed the day it took effect. No manual cleanup is required on your end: GitHub says the retired models are removed automatically. The one thing that might need action is on the admin side — Copilot Enterprise and Copilot Business organizations may need to turn on the replacement models through their model policies before anyone can select them.

What changed on October 2, 2026

GitHub first flagged the change in a changelog post published September 3, 2026, titled "Upcoming deprecation of selected GitHub Copilot models". It said GitHub would "deprecate the following models across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions) on October 2nd, 2026," and named four models: Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7.

On October 2 itself, GitHub published a follow-up post, "Selected models in GitHub Copilot deprecated," confirming the change had gone live: "As of today, October 2, 2026, we have deprecated the following models across all GitHub Copilot experiences." Both posts list the same four retired models and the same effective date, so this is one of the more cleanly executed deprecation cycles GitHub has run — the plan matched the outcome.

This is also not an isolated event. GitHub has been retiring Copilot models on a rolling, roughly monthly cadence throughout the second half of 2026. An August 31 changelog post, for instance, covered a separate wave that took Gemini 3.1 Pro, Claude Opus 4.5, Claude Opus 4.6, Claude Sonnet 4.5, Claude Sonnet 4.6, and Raptor Mini out of rotation on September 1. If you manage Copilot for a team, the October 2 wave is best read as one checkpoint in a continuing churn of the model lineup, not a one-off.

That cadence is worth internalizing if you're responsible for keeping Copilot usable for a team: roughly every three to five weeks, GitHub has been publishing a new "upcoming deprecation" notice, giving roughly a month of lead time before the model actually disappears, followed by a same-day confirmation post once the change lands. Treating each pre-announcement as a to-do item — rather than waiting for the confirmation post — is the only way to avoid a workflow quietly breaking on the effective date.

The four deprecated models and their replacements

Here is the full list from GitHub's own changelog, with the deprecation date and the suggested alternative for each model:

Both Gemini Flash models consolidate into a single replacement, Gemini 3.8 Flash, and Kimi K2.7 Code's successor is Kimi K3. The Claude Opus row is the one worth reading twice — see the next section for why.

Why GitHub's own posts disagree on the Claude Opus replacement

Here's a detail that's easy to miss if you only read one of GitHub's two posts: the September 3 pre-announcement lists Claude Opus 5 as the suggested alternative to Claude Opus 4.7. The October 2 confirmation post, published the day the deprecation actually took effect, lists the alternative as Claude Opus 5.5 instead. The table rows are otherwise identical — same model, same deprecation date — only the suggested replacement changed between the two posts.

GitHub's own supported models documentation lists both Claude Opus 5 and Claude Opus 5.5 as generally available options with the extended 1-million-token context window, so this isn't a typo pointing at a nonexistent model — both exist in Copilot's current lineup. The practical read is that GitHub updated its guidance in the four weeks between the two posts, and the October 2 confirmation is the more current recommendation. If you're migrating off Claude Opus 4.7 today, default to Claude Opus 5.5 and treat "Claude Opus 5" as the earlier, now-superseded suggestion.

Which Copilot surfaces this touches

GitHub's language is deliberately broad: the deprecation applies across "all GitHub Copilot experiences," and both changelog posts spell out the same five surfaces by name:

• Copilot Chat
• Inline edits
• Ask mode
• Agent mode
• Code completions

In practice that means the model picker in VS Code, the Copilot Chat panel on github.com, the Copilot CLI, and the GitHub Copilot mobile app are all affected if you had explicitly selected one of the four retired models rather than leaving Copilot on its automatic model selection. If you use Copilot's agent mode for longer autonomous tasks — the kind of workflow covered in our look at Copilot's computer-use features — it's worth checking which model that agent was pinned to, since agent-mode configurations are easy to set once and forget.

It's also worth distinguishing between users who rely on Copilot's "Auto" model selection and those who pin a specific model by name. Auto selection routes each request to whichever model GitHub currently considers the best fit, and GitHub manages that routing centrally, so those users were never going to see a retired model appear regardless of the October 2 change. The deprecation primarily affects anyone who deliberately chose Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, or Claude Opus 4.7 from the model dropdown, or any integration, extension, or automation that references one of those four model names directly in its configuration or API calls.

What Enterprise and Business admins need to do

For individual Copilot users on a personal plan, there's generally nothing to configure — GitHub's default model enablement covers the replacement models. The administrative step applies specifically to organizations running Copilot Enterprise or Copilot Business with custom model policies. GitHub's changelog puts it this way: "Copilot Enterprise and Copilot Business administrators may need to enable access to the alternative models through their model policies in Copilot settings."

The process GitHub describes is straightforward: an admin opens their organization or enterprise Copilot settings, goes to the Models tab, and confirms the policy is enabled for the specific replacement model (Gemini 3.8 Flash, Kimi K3, or the relevant Claude Opus version). Once that policy is turned on, GitHub says, "users will see the model in the Copilot Chat model selector in supported GitHub Copilot experiences." GitHub's documentation on managing enterprise policies for Copilot covers where these settings live and who can change them — enterprise owners can set policy enterprise-wide or delegate the decision to individual organization owners.

If your team pins a specific model inside editor configuration — for example in a VS Code settings file or a Copilot extension config — that pinned reference also needs updating, since a policy being enabled doesn't retroactively fix a hardcoded model name in a config file that still points at a retired model.

Do you need to manually remove anything?

No. Both changelog posts include the same reassurance, worded almost identically: the September post says "No action is required to remove the models after they have been deprecated," and the October 2 confirmation repeats it as "No action is required to remove the deprecated models." GitHub handles the removal on its end; a retired model simply stops appearing as a selectable option. The only follow-up work is making sure any workflow, script, or CI configuration that referenced a retired model by name is updated to reference its replacement, since an automation that explicitly calls for "Claude Opus 4.7" will fail once that model is gone, rather than silently falling back to something else.

Neither post specifies exactly what happens to an in-progress session or a saved chat history that used one of the retired models — GitHub's wording addresses the model's availability going forward, not historical conversations. If precise behavior for saved sessions matters for your workflow, the safest approach is to check GitHub's support channels directly rather than assume, since that detail isn't covered in either changelog post.

The next wave: more deprecations on October 19

If you're updating configurations now, it's worth doing it once and covering the next round too. A separate GitHub changelog post, published September 18, 2026, announces a second deprecation wave for October 19, 2026, covering six more models: Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5. The suggested replacements are Gemini 3.8 Flash, GPT-5.6 Sol, GPT-5.6 Sol, GPT-5.6 Luna, GPT-5.6 Luna, and Grok 4.6, respectively. GitHub's wording on admin action and cleanup for that wave matches the October 2 wave almost verbatim, so the same enable-the-policy, no-manual-cleanup pattern applies.

How to check which models you're actually using

Before assuming you're unaffected, it's worth confirming which model Copilot has been serving you. A few places to check:

• VS Code: open the Copilot Chat model picker and see whether a retired model name is still shown as selected, or whether it has already fallen back to "Auto." Our recent coverage of the VS Code 1.140 Copilot harness update walks through the current model-selector UI if the layout looks unfamiliar.
• github.com Copilot Chat: the same model dropdown appears in the web chat interface.
• Organization settings: admins can check the Models tab under Copilot settings to see which policies are currently enabled and which are off by default.
• CI/CD and scripts: grep any Copilot CLI invocations, GitHub Actions workflows, or extension configs for the literal strings "Gemini 3.5 Flash," "Gemini 3.6 Flash," "Kimi K2.7 Code," or "Claude Opus 4.7."

If you're also comparing how Gemini's current lineup stacks up against other assistants for day-to-day use outside Copilot, our ChatGPT vs. Gemini comparison covers pricing and feature differences between the two ecosystems.

What next

The immediate list is short. If you or your team explicitly selected Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, or Claude Opus 4.7 anywhere in Copilot, that selection no longer resolves to a working model as of October 2, 2026 — switch to Gemini 3.8 Flash, Kimi K3, or Claude Opus 5.5 respectively. If you're an Enterprise or Business admin, check the Models tab in your Copilot policy settings and confirm the replacement models are actually enabled rather than assuming default enablement covers you; GitHub's own wording is "may need to," not "will," which means some organizations are already covered and others aren't. Finally, mark your calendar for October 19, 2026, when the second wave retires six more models including two full GPT-5.x tiers and a Grok version — the same update pass that fixes your October 2 references can cover that wave too. For the authoritative, most current model list at any time, GitHub's own supported models page is the page to bookmark rather than relying on any single changelog post, including this summary of it.

Written byFelix Moreau

Felix Moreau writes Pandromeda's software coverage and how-to guides. He covers Windows, macOS and Linux updates, the apps people rely on, emulators and developer tools, and turns official documentation into clear, numbered steps that work on the current version.

More from Felix MoreauAll Software & Guides coverage