GitHub Copilot Model Retirements: What to Check Before December 14

A model you choose from the GitHub Copilot picker can also be named in a team guide, a reusable workflow or an integration. Changing your own selection handles only one of those references. Before December 14, find where your team relies on a retiring model and try a representative task with a replacement you can actually access.

An illustrated calendar, blue geometric modules and a paper checklist on a desk.
AI-generated concept illustration of planning a model change.

In its November 13 announcement, GitHub listed Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3 for deprecation on December 14, 2026, across Copilot experiences. GitHub says Claude Haiku 4.5 will remain available for individual subscribers on annual plans. The announcement did not name a replacement for Kimi K3.

Find the model references you rely on

Start with the model you deliberately select for recurring work. Then check your team’s instructions and any integration you maintain that selects a model by name. For each match, write down where it is used, who can change it and what a successful result looks like. A training document with an old screenshot needs a different update from an integration that requires a specific model identifier.

Confirm the replacement in the same Copilot experience

GitHub recommends Gemini 4 Hydrogen for Gemini 3.8 Flash and Claude Fable 5.2 for Claude Fable 5, among other pairings in the announcement. Open the picker in the experience you use and verify that your account can select the candidate. GitHub says Business and Enterprise customers can receive suggested models through default enablement, subject to administrator settings. For cloud agent sessions, GitHub’s model selection guide explains where you can choose a model and where Auto applies.

Review a representative task before updating the whole workflow

Suppose your team uses Copilot to summarize small pull requests. Pick a known example in an approved workspace and give the replacement the same instructions. Check whether the summary identifies the behavior change, cites the affected files and distinguishes test evidence from assumptions. Keep the review criteria fixed so a difference in the prompt does not obscure what changed.

For a task that proposes edits, use a disposable branch and inspect the diff before accepting anything. Record the Copilot experience, selected model and result. If the output misses a required detail, adjust the workflow and try another representative example. TechPulse’s pull request review guide covers how to follow checks to their evidence.

Update the instructions someone will use next week

Change the reusable instructions or integration only after checking access and reviewing the output. Note who owns the switch and where teammates can get help if the replacement is missing. For a model without a named successor in GitHub’s notice, make the choice explicit and review it against the same task criteria. Keep the date and result in the handoff so the next person can repeat the check.

Prepared with AI assistance using GitHub’s announcement and documentation. The workflow examples are illustrative. Read the TechPulse editorial policy.

MD Rafikul Islam

Written by

MD Rafikul Islam is a software developer and editor of TechPulse. He writes about developer tools, hardware, AI, and practical technology decisions. Some articles are based on cited documentation and analysis rather than hands-on testing; readers should check each article for sources and testing disclosures. Corrections are welcome at rony.yf25@gmail.com.

✍️ Leave a Comment

Your email address will not be published. Required fields are marked *