Field SOP
Field SOP

GitHub Copilot: Six Models Deprecated December 14, What to Fix

Six GitHub Copilot models are deprecated on December 14, 2026, and the official November 13, 2026 changelog gives six migration sentences but no configuration file name, model ID string or migration script. This SOP turns those sentences into a checkable list: replace hardcoded model names in workflows and integrations, enable model policies in Copilot settings, and understand what global default, the model selector, Copilot Chat, inline edits and code completions each do. All six model names and suggested alternatives are quoted as published, Kimi K3 stays To be announced, and the Claude Haiku 4.5 annual-plan exception for individual subscribers is spelled out.

Published November 16, 20269 min read
<!-- github-copilot-model-deprecation-migration-sop | sop | GitHub Copilot: Six Models Deprecated December 14, What to Fix -->

GitHub Copilot: Six Models Deprecated December 14, What to Fix. Read the official changelog twice and the puzzle stays the same: it hands you a date, a list, a column of suggested alternatives, and six short sentences about what to do, but not one configuration file name, not one model ID string, not one migration script. This SOP stays strictly inside what that page supports. It takes the six official migration statements and turns them into a checklist you can follow and verify: which switches an administrator touches, what an ordinary user sees, and what nobody needs to do once the models are retired. As of November 15, 2026, the changelog is the source this post relies on throughout, and this post adds no detail the page does not give.

What the November 13, 2026 changelog actually says

Pin down the notice first. The GitHub Copilot changelog page is dated November 13, 2026, marked Retired, and labeled a one minute read. Its scope line is worth quoting in full: "We will deprecate the following models across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions) on December 14, 2026:" The effective date is December 14, 2026. Some aggregators date this news to November 14, but the official changelog says November 13, 2026, so that is the date used here, with no attempt to explain the one day difference. This post organizes what the changelog can verify. It does not invent configuration details, model identifiers, migration scripts, or plan and price differences that the page never lists.

Why What to Fix is the hard part

Why focus on What to Fix instead of assuming the system will switch everything over? Because of the six official sentences, only one truly asks a human to change something, and it is the one most easily missed. The automatic enablement described later is an organization level switch that makes replacement models available; the official text never says it will rewrite a hard-coded model name sitting inside a workflow or an integration. Confuse those two and you get the classic deadline failure: an administrator believes "the alternatives are automatically enabled, so we are done," while a pipeline still calls a name that has been deprecated. Splitting the work into what administrators do, what users see, and what nobody needs to do after deprecation is exactly what prevents that mismatch.

The six deprecated models and their suggested alternatives

The six models and their suggested alternatives are copied row by row from the official table, with no model name changed and no replacement filled in on the official page's behalf.

ModelDeprecation dateSuggested alternative
Claude Fable 52026-12-14Claude Fable 5.2
Claude Haiku 4.5*2026-12-14Claude Haiku 5.5
Gemini 3.8 Flash2026-12-14Gemini 4 Hydrogen
Gemini 4 Argon2026-12-14Gemini 4 Carbon
Grok 4.62026-12-14Grok 4.7
Kimi K32026-12-14To be announced

* Claude Haiku 4.5 will remain available to individual GitHub Copilot subscribers on annual plans. The deprecation of Claude Haiku 4.5 does not apply to these customers.

Three details deserve care. First, the suggested alternative for Kimi K3 is To be announced, which means it has not been published; do not fill in a model name on the official page's behalf, and do not turn that cell into a claim that there is no replacement. Second, Gemini 4 Hydrogen appears here only as the suggested alternative for Gemini 3.8 Flash; this post gives it no section of its own, no evaluation, and no extra detail. Third, all six deprecation dates are identical, 2026-12-14, with no staging, no gradual rollout, and no immediate removal. Keep those three points straight and the six steps start from the right place. The list, the date, and the suggested alternatives are fully specified; everything beyond them is not.

Six steps, one official sentence each

Each step below matches one official sentence. For each, you get the sentence it rests on, the action to take, what to check, and the most common misreading. Follow them in the order the changelog gives them.

Step 1: Update hard-coded model names in workflows and integrations

Official sentence: "Please update your workflows and integrations to use the supported models before this date." The action is concrete: go through your workflows and integrations, find every hard-coded old model name, and replace it with the suggested alternative from the table above, completing the work before December 14, 2026. What to check: search for each of the six old names and confirm none still appears in your workflows and integrations, and that every replacement falls within the supported models. The common misreading is to assume that enabling a replacement in settings makes the old name in a script expire or update on its own. The official sentence puts the update on you, gives no promise of rewriting hard-coded strings, and names no configuration file or path, so the audit can only be done by asking where model names appear, not by inventing a file to edit.

Step 2: Decide whether your organization is under default model enablement

Official sentence: "Under default model enablement, the suggested alternatives are automatically enabled for Copilot Enterprise and Copilot Business customers unless an administrator has turned off the global default or explicitly disabled the model." This step is about classification, not clicking switches. Determine whether your organization is on Copilot Enterprise or Copilot Business, whether an administrator has touched the global default, and whether any model has been explicitly disabled. What to check: confirm the subscription type, the current state of the global default, and that no explicit disable record exists for the replacement models. The common misreading is to read this as "every user switches automatically, nothing to do." The official wording is narrower: automatic enablement covers Copilot Enterprise and Copilot Business customers, and applies when an administrator has neither turned off the global default nor explicitly disabled the model. Individual subscribers, and organizations that turned off default enablement, are outside what this sentence describes.

Step 3: If the global default is off, enable access through model policies

Official sentence: "If you've turned off the global default, you can enable access to the alternative models through their model policies in Copilot settings." The action: go into Copilot settings, find the model policies for each model, and turn on the replacement models listed in the table. What to check: compare against all six rows and confirm every replacement model is enabled in model policies, with no row skipped, especially when there is more than one replacement target. The common misreading is to hunt for a migration entry point that does not exist. The official sentence names model policies; it gives no configuration file name, no path, no command, and no migration script.

Step 4: Verify visibility in the Copilot Chat model selector

Official sentence: "Once enabled, users will see the model in the Copilot Chat model selector in supported GitHub Copilot experiences." This step is verification, not configuration. Have a real user open the Copilot Chat model selector and confirm each replacement model now appears in the list. What to check: sample the in-scope experiences and confirm the replacement models are visible and selectable, checking a second entry point if needed. The common misreading is to treat "enabled by an administrator" as "visible to users" and skip the check. The official text casts enablement and visibility as two successive steps, with visibility as the result of enablement, so it is worth verifying on its own.

Step 5: No manual removal is needed after deprecation

Official sentence: "No action is required to remove the models after they have been deprecated." The action here is to do nothing: schedule no cleanup task, delete no entry, write no removal script. What to check: confirm nobody on the team has put "remove the old models" on the deadline to-do list, so no needless work gets done. The common misreading is to treat deprecation as a signal that someone must pull the old models out of some place by hand. The official line says no action is required; it is the least effort of any item in this post and the one most likely to be over-engineered.

Step 6: Enterprise customers with questions contact their account team

Official sentence: "GitHub Enterprise customers with questions or concerns are encouraged to contact their account team for assistance." This step supplies the escalation path. If the first five steps leave something unclear, or your organization does not match the default assumptions, go to the account team instead of guessing. What to check: confirm someone in the organization knows this channel and that open questions have an owner. The common misreading is to treat media arithmetic about plans, credits, or quotas as an official answer and act on it. This post carries no price, plan difference, or quota conversion; open questions go to the account team, as the official sentence directs.

What is verifiable and what is not

After the six steps, it is worth restating the boundary. As of November 15, 2026, the changelog verifies a definite set of things: a list of six deprecated models, one effective date of December 14, 2026, a suggested alternative for each row, the scope across GitHub Copilot experiences, the enablement path through default model enablement, the global default, and model policies, the visibility point in the Copilot Chat model selector, the statement that no removal action is needed, and the account team channel for Enterprise customers. Equally definite is what the page does not give: any specific configuration file name or path, any model ID string, any migration script or command, any price or plan difference, and the not-yet-published replacement for Kimi K3. Media arithmetic and plan comparisons stay out. Separating "given" from "not given" means every step above lands on one official sentence and can be checked without outside guesswork; where the page is silent, say it is silent or say To be announced rather than supplying a plausible answer. That habit is the whole reason this post refuses to fill in details the changelog never wrote down.

A few posts on this site already cover migration and tooling, and this one takes a different angle, so do not blend them. The API version migration SOP deals with upgrading one model's API version; this post deals with enablement and verification after models are deprecated on the Copilot side. The model sunset cost post calculates the cost of retiring a model; this post carries no price or quota. The GitHub AI tools metadata comparison is a metadata review; this post compares no tool capabilities and runs no benchmarks. Each stands on its own object; this one stays on the six migration sentences of the GitHub Copilot changelog.

FAQ

Q1: What happens on December 14 if I changed nothing? A1: The official notice asks you to update model names in workflows and integrations before that date, so any pipeline with a hard-coded old name needs your own replacement. Automatic enablement of the alternatives covers Copilot Enterprise and Copilot Business customers when an administrator has not turned off the global default, and it does not rewrite hard-coded names in a script. Change nothing and the risk sits exactly in those hard-coded places.

Q2: The alternatives are automatically enabled, so does every user need to do nothing? A2: No. The official wording is limited to Copilot Enterprise and Copilot Business customers, and applies when an administrator has neither turned off the global default nor explicitly disabled the model. An organization that turned off the global default must enable access through model policies in Copilot settings as in Step 3; other cases are outside what that sentence describes.

Q3: Which model should Kimi K3 move to? A3: The suggested alternative says To be announced, meaning it has not been published. This post does not fill in a model on the official page's behalf, and it does not turn that cell into "there is no replacement." What you can do before the deadline is record the official wording as written and update workflows and integrations once an alternative is published.

Q4: Do I need to find a configuration file and delete the old model names? A4: No, and the official page gives no configuration file name or path at all. What you do is update hard-coded model names in workflows and integrations, enable the alternatives through model policies in Copilot settings, and verify visibility in the Copilot Chat model selector. After deprecation the official text also says no action is required to remove the models.

Q5: When was this announced, and when does it take effect? A5: The official changelog page is dated November 13, 2026, and the effective date is December 14, 2026. Some aggregators date the news to November 14; this post follows the official November 13, 2026. The time anchor for this post is As of November 15, 2026.

Related reading

This article is AI-assisted and human-edited. Last updated: 2026-11-16

FAQ

What happens on December 14 if I changed nothing?
The official notice asks you to update model names in workflows and integrations before that date, so any pipeline with a hard-coded old name needs your own replacement. Automatic enablement of the alternatives covers Copilot Enterprise and Copilot Business customers when an administrator has not turned off the global default, and it does not rewrite hard-coded names in a script. Change nothing and the risk sits exactly in those hard-coded places.
The alternatives are automatically enabled, so does every user need to do nothing?
No. The official wording is limited to Copilot Enterprise and Copilot Business customers, and applies when an administrator has neither turned off the global default nor explicitly disabled the model. An organization that turned off the global default must enable access through model policies in Copilot settings as in Step 3; other cases are outside what that sentence describes.
Which model should Kimi K3 move to?
The suggested alternative says To be announced, meaning it has not been published. This post does not fill in a model on the official page's behalf, and it does not turn that cell into "there is no replacement." What you can do before the deadline is record the official wording as written and update workflows and integrations once an alternative is published.
Do I need to find a configuration file and delete the old model names?
No, and the official page gives no configuration file name or path at all. What you do is update hard-coded model names in workflows and integrations, enable the alternatives through model policies in Copilot settings, and verify visibility in the Copilot Chat model selector. After deprecation the official text also says no action is required to remove the models.
When was this announced, and when does it take effect?
The official changelog page is dated November 13, 2026, and the effective date is December 14, 2026. Some aggregators date the news to November 14; this post follows the official November 13, 2026. The time anchor for this post is As of November 15, 2026.

Related

Field SOP

Yuanqi Is Down: Why Your Legacy Workflow Can't Be Exported?

Tencent Yuanqi formally shut down on November 9, 2026: application access, data read/write, API calls and the console are all closed, and agents published to Official Accounts, service accounts or WeChat Customer Service no longer respond. Every figure in this checklist comes from four official pages (discontinuation notice, data-export guide, ADP migration guide, refund guide), arranged as a five-step salvage plan: new-agent creation and API key issuance stopped October 29; before shutdown, self-service export paths cover conversation records, knowledge bases, workflows and Widgets with per-space one-click export; after shutdown and until January 9, 2027 the data-export console still delivers backups, after which all user data is deleted; migration targets the officially named Tencent Cloud Agent Development Platform (ADP), where knowledge bases, new-version workflows and Widgets import directly, but the official guide states verbatim that legacy Yuanqi workflows cannot be exported and must be rebuilt by hand, plugins must be recreated as custom tools, and the two platforms' official models and plugins differ enough that answers may change; refunds cover only unconsumed Hunyuan prepaid resource packs bought on Tencent Cloud for legacy API calls, with post-paid billing closed first as officially warned to affect other post-paid services, and 5-10 business days for a refund after review.

Nov 10, 20269 min read
Field SOP

Bedrock AgentCore V2 Snapshots: Four Patterns, One Pre-warm

Turns the four official AWS section names into a checkable sequence for Bedrock AgentCore runtime V2 session startup: measure Latency and OverheadLatency in CloudWatch first, move imports and static data into init, use Working with hooks for build-time large files, then send a background warm-up request with Pre-warm on user engagement and keep the session ID. It records the 120-second /ping gate, that the V2 default applies only to new and updated runtimes, that existing ones need explicit platformVersion=V2, and that the sample code is for non-production use. The Region and cold-start discrepancies between the two official pages are kept side by side.

Nov 15, 20269 min read