GitHub Copilot will retire six models on December 14

Test the replacement in the same mode and repository where the old model works.

GitHub says it will remove six models from Copilot on December 14, 2026. If a team depends on one for a repeatable task, now is the time to verify the suggested replacement is enabled and test the same workflow with it. The notice covers Copilot Chat, inline edits, ask and agent modes, and code completions. GitHub’s retirement notice

Which models are changing

Retiring model	GitHub’s suggested replacement
Claude Fable 5	Claude Fable 5.2
Claude Haiku 4.5	Claude Haiku 5.5
Gemini 3.8 Flash	Gemini 4 Hydrogen
Gemini 4 Argon	Gemini 4 Carbon
Grok 4.6	Grok 4.7
Kimi K3	Pending in GitHub’s notice

GitHub says Claude Haiku 4.5 will remain available to individual Copilot subscribers on annual plans. For Kimi K3, GitHub had not yet identified a replacement in the November 13 notice. Teams that depend on it can test the alternatives they already have and check GitHub’s notice again before making a permanent change.

Verify access with an ordinary team account

For Copilot Business and Enterprise, GitHub says the suggested alternatives are enabled under its default model setting unless an administrator disabled that setting or the specific model. If the team uses custom policies, check the replacement in Copilot settings. Then have someone with a normal developer seat open the model selector in the client the team uses. That catches a model enabled in one organization or client but missing where the work happens. GitHub’s Copilot model documentation

A separate November 13 notice says Gemini 4 Hydrogen is generally available in Copilot with a gradual rollout. GitHub identifies eligible plans and notes the policy requirement for Business and Enterprise. The GitHub rollout details explain access in Copilot. For developers calling Google directly, see MustHave’s guide to the Gemini 4 Hydrogen API.

Run the same coding task twice

Choose a known task in the same repository, such as fixing a failing test with a small patch. Keep the prompt, baseline commit, test command, and mode fixed. Try the old model and its replacement, then compare whether they pass the tests, the size of the diff, and which files or commands each one attempts to touch.

Run at least one check in the mode your team uses most. An acceptable chat answer tells you little about what an agent may edit or run. Once the replacement passes the team’s checks, revise the internal instructions and workflow configuration that name the old model.