Coding with AI

Six Copilot Models Retire on December 14

GitHub will retire six Copilot models on December 14. See the replacements, Haiku exception and model-policy checks to make before switching.

Six glass model tiles beside a separate replacement tile on a navy background.

GitHub announced on November 13 that it will retire six AI models across GitHub Copilot on December 14, 2026. The change covers Chat, inline edits, ask and agent modes, and code completions. Five models have named suggested alternatives; GitHub has yet to name one for Kimi K3. If a team selects a model explicitly, it should check that workflow before the deadline. The announcement does not give a cutoff hour or time zone.

Six models reach the December 14 deadline

GitHub’s retirement announcement lists these pairs. “Suggested alternative” is GitHub’s recommendation, not evidence that a saved model choice or integration will change automatically.

Model scheduled for retirement GitHub’s suggested alternative December 14 action
Claude Fable 5 Claude Fable 5.2 Check the replacement’s organization policy before changing a saved selection.
Claude Haiku 4.5 Claude Haiku 5.5 Check whether the individual annual-plan exception applies.
Gemini 3.8 Flash Gemini 4 Hydrogen Check the selected client, plan and model policy.
Gemini 4 Argon Gemini 4 Carbon Check the selected client, plan and model policy.
Grok 4.6 Grok 4.7 Check the selected client, plan and model policy.
Kimi K3 To be announced Wait for an official Copilot alternative before choosing one.

All six rows have the same announced date. GitHub’s supported-models reference still lists the outgoing models as available at the time of this reading. The dated announcement supplies the future deadline. This is a Copilot availability change; it says nothing about whether the model makers will continue offering the same models through their own APIs.

A replacement shown in GitHub’s table may not appear in every person’s model picker. GitHub says availability depends on the Copilot plan and the place where a person uses it, such as GitHub.com or an IDE. Its documentation also lists minimum IDE and extension versions for recent models, with some values still marked “TBD.” Check the actual client before treating a model as a usable replacement. For model details outside Copilot, the AI model tracker is useful context, but Copilot’s announcement and settings decide what is available inside Copilot.

Check the Haiku annual-plan exception

The announcement makes one explicit exception: Claude Haiku 4.5 remains available to individual GitHub Copilot subscribers on annual plans. GitHub says the December 14 deprecation does not apply to those customers. It does not give an end date for that exception.

That wording is narrow. An individual annual subscriber can verify the subscription and keep using Haiku 4.5 under the stated exception. A monthly subscriber or an organization should not assume the exception covers their account. For those users, GitHub suggests Claude Haiku 5.5. The Copilot plan comparison gives broader buying context, but the exception comes from GitHub’s November 13 notice, not from plan pricing.

Default enablement needs an admin check

The announcement says suggested alternatives are enabled automatically for Copilot Business and Enterprise customers under default model enablement, unless an administrator has disabled the global default or that model. It also says administrators can enable alternatives in model policies when the global default is off.

There is a material qualification in GitHub’s supported-models documentation: Claude Fable 5, 5.1 and 5.2 are outside the default enablement policy and disabled by default regardless of the “Default availability” setting. The same section excludes open-weight models such as Kimi K3, pre-GA models and models barred by applicable data residency or FedRAMP restrictions. The broad wording of the announcement therefore does not establish that Fable 5.2 will appear automatically in every Business or Enterprise account. An administrator should inspect the model’s actual policy and the intended user’s model picker.

This matters most for teams that have chosen Fable 5 explicitly. GitHub recommends Fable 5.2, but an available replacement in a reference table is different from one enabled for a particular organization. A policy owner can confirm whether Fable 5.2 is permitted, whether its data handling fits the organization’s requirements, and whether the intended Copilot client offers it. The public documents do not verify any individual organization’s configuration.

Make a small migration worksheet

A team can inventory each place where it deliberately picked one of the six models: a Chat habit, an agent workflow, an IDE setting, or an integration it maintains. The table above identifies the announced replacement; it does not tell the team whether a specific task will give acceptable results. Record the checks so a missing model or changed output is visible before December 14.

Record for each workflow Why it helps
Current model and Copilot client Identifies the selection that needs attention and where it runs.
Plan and policy owner Shows who can verify availability and enablement.
Proposed replacement Keeps GitHub’s recommendation separate from the team’s final choice.
Representative task and result Lets the team compare an output it actually cares about.
Usage or cost observation Captures a measured effect instead of assuming a saving.
Decision and owner Makes the next change explicit.

These are recommended checks, not results of a test we performed. Choose a task that the workflow already handles, run it with a permitted replacement, and review the output before updating a saved model selection. If that replacement is unavailable, check the supported client and plan tables and ask the policy owner to inspect the setting. For plan selection, the AI plan finder can organize the broader choice, though it cannot confirm an organization’s model policy.

Kimi K3 needs a different line in the worksheet: “replacement not announced.” GitHub has not named its Copilot alternative in this notice. Do not treat a newer Kimi model listed in a general model tracker as GitHub’s recommendation. Enterprise customers with a concern can follow GitHub’s suggestion to contact their account team. For users considering the suggested Gemini models, our separate articles on Hydrogen and Carbon discuss direct API pricing; those API figures do not establish what a Copilot subscription will bill for a migration.

GitHub says no action is needed to remove the old models after deprecation. The action before the date is to find deliberate model choices, confirm a permitted alternative and test the workflows that matter to the team.

Sources

All pages read on November 16, 2026.

Free toolIs a monthly AI plan cheaper than paying per use?Compare for my usage
DailyTech Editorial Team

We explain AI with numbers you can check. Every price, limit and benchmark links to its source and the date we read it. How we work · Report an error