Public developments and original analysis · Announcement date: · Sources
GitHub announced on November 13 that six models will leave multiple Copilot experiences on December 14, including chat, inline edits, ask and agent modes, and code completion. Development leaders now have a window to identify teams that actually depend on those choices and arrange replacements before failures reveal overlooked dependencies.

The affected models are Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3. Suggested successors are Fable 5.2, Haiku 5.5, Gemini 4 Hydrogen, Gemini 4 Carbon and Grok 4.7; the Kimi K3 replacement remains unannounced. This is Copilot’s product schedule, not evidence that the vendors’ own APIs stop on the same day. The exception retaining Haiku 4.5 for individual annual subscribers does not extend to enterprise teams.
Default availability and organization policies
GitHub says suggested alternatives are generally enabled for Business and Enterprise customers under default model enablement. A disabled global default or explicit model restriction requires action through model policies. A recommended name and permission to use it in an organization are separate facts. A model visible in a developer’s personal account does not establish availability in a company account.
The supported-model documentation also excludes categories from default availability, including pre-GA models, open-weight models and models outside the relevant data-retention agreement. It currently lists the Fable family in the last category. The announcement’s general rule therefore needs a model-specific check against current labels and organization policy. Teams should not assume every row is automatically approved or remove global restrictions merely to meet the deadline.
Consider a hypothetical company that permits a model for public code but applies different requirements to repositories containing internal business rules. Checking only the chat selector misses the records, accounts and surfaces involved. A useful migration record brings together the team, task, current selection and permitted replacement, leaving unresolved questions with the policy owner.
There is a tradeoff: default availability reduces waiting for individual administrator actions, while explicit approval supports decisions tied to company data requirements. Existing procurement and data-handling rules should determine the choice. The retirement creates a reason to organize dependencies, not a requirement to build a new approval platform. Existing configuration records and clear ownership may be sufficient.
Following the team’s actual entry points
The announcement covers several Copilot experiences. Changing the chat selection should be followed by checking existing editor, agent and integration choices. GitHub’s documentation distinguishes availability by feature and surface. Walk through the team’s existing work rather than treating a new name in one selector as completion for every user.
In the hypothetical team, one developer asks chat to explain code, another uses an agent for multi-file changes and another mainly uses completion. Their useful outcomes differ: whether an explanation identifies limitations, a change passes project checks, or a completion follows the local code style. Recent real tasks with confirmed outcomes can provide a comparison without building a universal coding leaderboard first.
A suggested successor provides direction without guaranteeing identical prompts, latency, usage costs or edits. Nor does every task need the strongest option. A faster model that handles the existing scope may be more useful for simple work. Complex changes warrant attention to human corrections and the effort required after checks fail.
Keep task inputs and previous results alongside model names. Hold project versions, tool permissions and supplied context as consistent as possible. If the replacement simultaneously receives more context or different tools, an improved result cannot be attributed to the model change alone.
Pending replacements and migration completion
The notice names no successor for Kimi K3. Teams should not promise delivery using an unannounced product. Work that must continue after the deadline can be evaluated on currently permitted alternatives. If results fall short, narrow automatic editing and keep manual work or established scripts available. Monitoring announcements and preparing a usable fallback can proceed together.
Migration completion also needs a practical definition. Administrator enablement establishes access readiness; one request establishes that one entry point responds. For common code changes, teams should verify project checks, human review and existing integrations continue to work. Saving the observed differences helps subsequent maintainers avoid repeating the same investigation across repositories.
Before December 14, prioritize dependencies that are active, shared or lack a replacement. An unused model does not need extra testing simply to complete a checklist; GitHub says users need not remove retired models themselves. The worthwhile work concerns how ongoing tasks will continue and who owns the differences already found.
Sources
- GitHub: upcoming December Copilot model deprecation
Source published: · Accessed:
- Supported AI models and default enablement exceptions
Accessed:
Mokyun Editorial · neo@mokyun.com
