GitHub Copilot Retires Six Models on December 14: Replacements and Admin Policy Checks

What GitHub is retiring​

GitHub's changelog entry of November 13 says the deprecation applies across all Copilot experiences. That covers Copilot Chat, inline edits, ask and agent modes, and code completions. All six models share the same date, and each has a suggested alternative:

Kimi K3 has no named replacement yet, so don't plan around a substitute GitHub hasn't announced.

The asterisk on Claude Haiku 4.5 is an exception. It stays available to individual Copilot subscribers on annual plans, and the deprecation does not apply to them. GitHub states no similar exception for any other model.

What GitHub says to do​

GitHub's guidance is short:

• Update workflows and integrations to supported models before December 14.
• Under default model enablement, the suggested alternatives are switched on automatically for Copilot Business and Copilot Enterprise. That holds unless an administrator turned off the global default or explicitly disabled the model.
• If you turned off the global default, enable the alternatives through their model policies in Copilot settings.
• Once enabled, a model shows up in the Copilot Chat model selector in supported experiences.
• You don't need to remove the deprecated models afterward.
• Enterprise customers with concerns should contact their GitHub account team.

Check the default-enablement claim before you rely on it​

The changelog says alternatives are enabled automatically. GitHub's default-availability documentation adds exceptions.

• The default policy only applies to models labeled "Delegate to Default Policy". That means models you haven't explicitly configured.
• Some models are disabled by default whatever the policy says. They include pre-GA models and open-weight models, and the docs name Kimi K3 as one of those.
• Models outside GitHub's data-retention agreement are also excluded. The docs list Claude Fable 5, 5.1 and 5.2 there.

So Claude Fable 5.2 is named as the replacement for Fable 5, yet the docs list it among models that are off by default. An administrator can't assume it is switched on. Check its status in your enterprise or organization model settings.

GitHub's supported-models page also has a footnote on the Fable models. Anthropic retains data, including prompts and outputs, by default to run safety classifiers. The page says customers can request zero data retention through the end of 2026 under a time-limited exemption. After that, continued use would require Anthropic's Enterprise Frontier Safeguards. Approval doesn't enable the model automatically. An administrator still has to turn it on.

For teams with strict data-handling rules, moving from Fable 5 to Fable 5.2 doesn't change that picture. It's worth raising with your account team now, before the deadline.

Practical checklist​

1. Inventory usage. Find where any of the six models are selected by name. That includes team defaults, documented workflows and scripts.
2. Pick replacements. Use GitHub's suggestions as the starting point. For Kimi K3, wait for GitHub's announcement.
3. Check policy status. In enterprise and organization model settings, confirm each replacement is Enabled, not Disabled or only "Delegate to Default Policy" under a default that is turned off.
4. Check client support. GitHub's docs say availability depends on your plan and client, such as GitHub.com or an IDE. Many minimum IDE versions are still marked "TBD", and the page calls them tentative. Test in the editors your developers actually use, and keep the IDE and Copilot extension current.
5. Test the behavior. Prompts tuned for one model may behave differently on another. GitHub's notice makes no promise of equivalent results.
6. Plan for Haiku 4.5 exceptions. Annual individual subscribers can keep it. Business and Enterprise users should plan for Haiku 5.5.

What the notice doesn't say​

GitHub's notice doesn't say:

• whether model identifiers in integrations will fail or fall back automatically after the date;
• whether in-progress sessions are affected;
• whether any settings migrate;
• whether the replacements perform comparably.

Those are open questions, so test rather than assume.

Not the first, and not the last​

This is routine for Copilot. GitHub's changelog shows earlier rounds. An Oct 2 notice retired Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code and Claude Opus 4.7. An Oct 19 round retired Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5.

A separate deprecation for Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra has been reported for November 16. That comes from a third-party model guide, not from GitHub's December notice.

Teams that hard-code model names will keep running into this. The practical fix is general: treat model choice as configuration you review regularly, and keep a documented fallback for each workflow. GitHub's docs warn that a model can be supported yet unavailable to a developer because the organization disabled it. That's the failure mode to test for. A model that exists in the picker for one person can still be blocked by policy for another.