Copilot drops six models on December 14, and Kimi K3 gets no named successor

GitHub, a Microsoft company, will deprecate six models across every GitHub Copilot experience on December 14, 2026. Kimi K3 is the only one for which GitHub has not named a replacement yet.

Six models leave on one date, with one exception

The removal covers Copilot Chat, inline edits, ask and agent modes, and code completions. According to GitHub’s changelog, the models and the alternatives it suggests are:

• Claude Fable 5, replaced by Claude Fable 5.2
• Claude Haiku 4.5, replaced by Claude Haiku 5.5
• Gemini 3.8 Flash, replaced by Gemini 4 Hydrogen
• Gemini 4 Argon, replaced by Gemini 4 Carbon
• Grok 4.6, replaced by Grok 4.7
• Kimi K3, with the alternative listed as to be announced

The Haiku entry carries a caveat. Claude Haiku 4.5 will stay available to individual Copilot subscribers on annual plans, and the deprecation does not apply to them. GitHub does not say why, or whether that carve-out has an end date.

For organizations, the changelog says the suggested alternatives are switched on automatically for Copilot Enterprise and Copilot Business under default model enablement. That does not happen if an administrator has turned off the global default or explicitly disabled the model. In that case, admins must enable the replacement through its model policy in Copilot settings, after which it appears in the Copilot Chat model selector. GitHub says nothing needs to be done to remove the old models once they are gone. It asks users to update workflows and integrations before the date, and tells Enterprise customers with concerns to contact their account team.

Kimi K3 users have no destination yet

Five of the six entries come with a direct successor, and in most cases it is a newer release from the same lab. Kimi K3 is different: the changelog lists no replacement, so anyone who pinned it in an integration has to wait for a follow-up or choose another model themselves. The source does not say when the replacement will be named, so teams have roughly a month to plan around that gap.

Gemini 4 Hydrogen, the suggested successor to Gemini 3.8 Flash, shipped the same day. GitHub’s separate changelog entry calls it generally available on Copilot Pro, Pro+, Max, Business and Enterprise, with a gradual rollout. It describes the model as built for fast agentic coding, tool use and multistep workflows, and says it follows instructions better than Gemini 3.8 Flash. Those are GitHub’s descriptions and have not been independently checked here. The entry does not give a premium request multiplier, and says billing follows provider list pricing under usage-based billing.

A tidy-up that tests how well teams pin models

The analysis angle here is cost and economics, and the economics are mostly about maintenance. Each model GitHub keeps in the picker is a model it has to serve, support and bill. Retiring older Flash-class and Haiku-class entries in favor of newer ones lets it concentrate capacity on the current generation. GitHub does not state that motive, so this is an inference from the pattern of the list, where nearly every row maps an older release to a newer one from the same vendor.

The billing change makes the choice of replacement more consequential. According to the Gemini 4 Hydrogen entry, Copilot now bills at provider list pricing under usage-based billing. If that applies to the other suggested alternatives too, which the deprecation notice does not say, moving from Claude Haiku 4.5 to Claude Haiku 5.5 or from Grok 4.6 to Grok 4.7 could change what a team pays, not just what it gets. Admins should check the models and pricing documentation GitHub links to before switching defaults.

Reliability is the other context. On the same day, GitHub reported elevated error rates on several GPT models in Copilot, between 19:05 and 21:27 UTC, which it attributed to an upstream provider, as we reported. The affected models were GPT-6.1 Terra, GPT-5.6 Sol, GPT-6 Luna and gpt-5.6-sol-fast, and GitHub advised users to switch models or use Auto. That incident is unrelated to the deprecation, but it shows why teams that hard-code a single model carry more risk than those that keep a fallback.

What the notice leaves unanswered

The changelog does not explain why these six were chosen, what usage they have, or whether the suggested alternatives cost the same as the models they replace. It does not say whether API or Copilot SDK users will see an error or an automatic redirect after December 14, only that no action is required to remove the models. The scope of the annual-plan exception for Haiku 4.5 is also thin: it names individual subscribers on annual plans and nothing else.

For now, the practical steps are the ones GitHub lists. Individual users will see the replacements in their model picker. Business and Enterprise administrators who turned off default model enablement should switch on Claude Fable 5.2, Claude Haiku 5.5, Gemini 4 Hydrogen, Gemini 4 Carbon and Grok 4.7 through model policies before the cutoff, and watch for a Kimi K3 follow-up.