GitHub Copilot's next model deprecation lands on 19 October 2026, when six models go away, and the replacements only switch on by themselves for some admins. If you turned off default model enablement, or explicitly disabled a model, nothing replaces it until you approve the successor. A first wave of four models already went on 2 October, so the same check applies today. For the wider picture of what model changes mean for engineering teams, see AI model releases and pricing: what changes for engineering teams.
Which Copilot models are being removed, and what replaces them
GitHub's September 18 changelog on mid-October deprecations names six models for removal on 19 October 2026. The October 2 changelog on deprecated models covers four more that are already gone.
| Deprecated model | Replacement | Removed on |
|---|---|---|
| Gemini 3.5 Flash | Gemini 3.8 Flash | 2 October 2026 (done) |
| Gemini 3.6 Flash | Gemini 3.8 Flash | 2 October 2026 (done) |
| Kimi K2.7 Code | Kimi K3 | 2 October 2026 (done) |
| Claude Opus 4.7 | Claude Opus 5.5 | 2 October 2026 (done) |
| Gemini 3.7 Flash | Gemini 3.8 Flash | 19 October 2026 |
| GPT-5.5 | GPT-5.6 Sol | 19 October 2026 |
| GPT-5.4 | GPT-5.6 Sol | 19 October 2026 |
| GPT-5.4 mini | GPT-5.6 Luna | 19 October 2026 |
| GPT-5 mini | GPT-5.6 Luna | 19 October 2026 |
| Grok 4.5 | Grok 4.6 | 19 October 2026 |
Four distinct replacements cover the second wave: GPT-5.6 Sol, GPT-5.6 Luna, Gemini 3.8 Flash and Grok 4.6. Those are the ones to put in front of whoever owns your model policy. Gemini 3.8 Flash replaces three removed models across both waves, which makes it the one most teams will touch first. Its pricing is covered in our Gemini 3.8 Flash pricing write-up.
Who has to act: default model enablement on or off
The September 18 changelog says that for Copilot Enterprise and Business, "Under default model enablement, the suggested alternatives are automatically enabled" unless an administrator has turned off the global default or explicitly disabled the model. That condition sorts admins into three cases.
If default model enablement is on and nothing is disabled, replacements are enabled automatically. They appear in the model selector in Chat, inline edits, ask and agent modes, and code completions. You don't need to do anything, though it's worth knowing which new providers just reached your developers.
If the global default is turned off, the replacement stays off until you enable it. Developers still on a removed model lose it on 19 October, and the successor isn't there to pick up the slack.
If an administrator has explicitly disabled the model, the changelog says the alternatives aren't automatically enabled. The sentence doesn't say whether "the model" means the retiring model or its replacement, so check the policy status of both.
The October 2 changelog states the admin duty plainly: "Copilot Enterprise administrators may need to enable access to alternative models through their model policies in Copilot settings." It also says "No action is required to remove the deprecated models." What a developer sees when selecting a removed model isn't described in either changelog. The likely result in the second and third cases is a model picker with neither the old model nor its replacement, but that's inference, not something GitHub states.
The default-enablement mechanics themselves are covered in GPT-6.1 Sol's Copilot on-by-default decision. Don't confuse any of this with the features policy starting 22 October, which is a separate setting covered in GitHub Copilot's default enablement policy for features.
How to enable the replacement models before 19 October
Enterprise and organization admins use different pages. Do the enterprise one first if you have an enterprise account, since delegating there decides whether org owners get a say.
1. Enterprise: set baseline model access
Per GitHub's docs on managing availability of default models, go to your enterprise from the Enterprises page, then click AI controls at the top of the page. In the sidebar, click Copilot, then Configure models. For each model, set the status to Enabled, Disabled or Delegate.
Pick Enabled for each replacement you've approved. Delegate lets organizations (or enterprise teams and apps) decide. Models you leave unconfigured show as "Delegate to Default Policy" and follow the "Default availability" policy: available to everyone if it's enabled, restricted if it's disabled.
2. Organization: edit the Models policy
Organization owners work from the docs page on managing Copilot policies. In the upper-right corner of GitHub, click your profile picture, then click Organizations. Select the organization and click Settings. In the sidebar, under "Code, planning, and automation", click Copilot.
Click Models. That page edits the policies that control availability of models beyond the basic models provided with Copilot, which may incur additional costs. For each policy, click the dropdown menu and select an enforcement option.
3. Check that the replacements appear
Once enabled, the replacements show in the Chat model selector in VS Code and on github.com, per the October 2 changelog. Open a Chat session in both and look for Gemini 3.8 Flash, Kimi K3 and Claude Opus 5.5 now, and GPT-5.6 Sol, GPT-5.6 Luna and Grok 4.6 before the 19th.
What this means for your AI tooling governance
The decision is a model-approval review with a deadline: do GPT-5.6 Sol, GPT-5.6 Luna, Gemini 3.8 Flash and Grok 4.6 go into your Copilot model policy before 19 October, or stay off?
The model makers differ: OpenAI (GPT-5.6), Google (Gemini), xAI (Grok), Moonshot (Kimi K3) and Anthropic (Claude Opus 5.5). That's background, not something the changelogs state. Whether each is acceptable under your data policy is your call. Neither changelog says anything about provider data terms, so check them yourself rather than assuming the old model's approval transfers.
Our judgment: if your org vets models, keep the global default off, but then put every new release on a recurring approval list. That's the standing-chore trade-off. Someone owns that list, or developers lose models on deprecation day with nothing behind them. For related coverage, see HighCircl's ai-news section, and for the API-side version of the same problem, how Sonnet 4.5's retirement was handled.
FAQ
Which GitHub Copilot models are deprecated on 19 October 2026?
Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5. Their replacements are Gemini 3.8 Flash, GPT-5.6 Sol (for GPT-5.5 and GPT-5.4), GPT-5.6 Luna (for GPT-5.4 mini and GPT-5 mini) and Grok 4.6.
Do I need to remove the deprecated models myself?
No. The October 2 changelog says "No action is required to remove the deprecated models." The only action is on the replacement side.
Will replacement models turn on automatically?
Only under default model enablement, and only for Copilot Enterprise and Business admins who haven't turned off the global default or explicitly disabled the model. In either of those cases, you enable the replacement yourself.
Where do admins enable replacement models?
At the enterprise level: AI controls, then Copilot, then Configure models. At the organization level: Settings, then Copilot, then Models. The steps above give the full path for each.
Does the deprecation change what Copilot costs?
Neither changelog states a pricing change, and neither mentions prices or multipliers at all, so there's nothing sourced to say either way. The Models policy page does note that models beyond the basic ones may incur additional costs.