GitHub Copilot’s November 16 model retirements: check the replacement before it disappears
Opus and Terra workflows face an immediate transition, with another retirement round due in December. Plan exceptions and administrator settings matter.
GitHub Copilot’s model menu changes on November 16, and another round of retirements is already scheduled for December. For developers, the immediate job is to check which model their workflow actually uses and whether their organization permits its replacement. A subscription alone does not answer either question.
The November retirement notice names Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra. GitHub recommends Opus 5.5 for the first two, and GPT-6.1 Sol or GPT-6 Luna for the OpenAI model. The change covers Copilot experiences including chat, inline editing, agent workflows and completions; it is a Copilot availability decision, not an announcement that those models disappear from every provider.
The exception that changes the checklist
Individual subscribers on annual plans retain access to Claude Opus 4.8 under the stated exception. That wording matters: it does not extend the exemption to Opus 5, to GPT-5.6 Terra, or automatically to an organization’s Business seats. Check the actual plan attached to the account doing the work.
For Business and Enterprise customers using default model enablement, GitHub says the recommended alternatives are enabled automatically unless an administrator has disabled the global setting or a particular model. Teams that maintain an allowlist therefore have an extra dependency: the replacement must be allowed before the developer can select it. GitHub removes retired options itself; the useful intervention is ensuring a supported route remains available.
A replacement name is not a migration test
GitHub’s supported-model documentation separates access by plan and client. A model visible on the website may not be available in the same way inside every editor. Check the tool where the work happens, including the installed extension or IDE version, rather than stopping at a successful web login.
There is also a practical distinction between compatibility and behavior. If a team uses an agent to prepare changes, a useful migration check is to give the replacement a familiar, bounded task: a small bug with an existing regression test, for example. Inspect the diff, the test result and the files it touched. That establishes whether the workflow still behaves acceptably without pretending one vendor-recommended substitution is a measured quality upgrade.
This becomes more consequential as Copilot moves into Slack and Teams pull-request workflows. The interface that starts a task can be far from the repository where its effects appear. Model policy and review rules need to follow that whole path.
December gives teams a second deadline
The November 13 announcement schedules six further retirements for December 14. It also contains a separate annual-plan exception for Claude Haiku 4.5, while Kimi K3’s recommended replacement is still to be announced. Those details should stay on the migration calendar rather than being guessed from naming conventions.
The productive response is a small inventory: the model selected in each workflow, the account policy that allows it, and a representative task that still passes review after switching. That is more durable than memorizing today’s menu. When the next name disappears, the team should already know which work depends on it.