GitHub Copilot retires 6 models on 14 Dec: Claude Fable 5, Gemini 3.8 Flash, Kimi K3
Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3 leave every Copilot surface on 14 December 2026; here is what to change.
· 7 min read
Written by AI from the sources linked below. How we use AI
On this page
GitHub will remove six models from every Copilot experience on 14 December 2026: Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3. It is the third retirement round in two months, and the pattern is now regular enough to plan around: if you pin a model by name in a workflow, a script or an organisation policy, you have until that date to move.
Key points
- GitHub announced on 13 November 2026 that six models leave all Copilot experiences on 14 December 2026: Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3.
- Suggested replacements are Claude Fable 5.2, Claude Haiku 5.5, Gemini 4 Hydrogen, Gemini 4 Carbon and Grok 4.7; for Kimi K3 the replacement is "to be announced".
- Claude Haiku 4.5 stays available to individual Copilot subscribers on annual plans; the retirement does not apply to them.
- On Copilot Business and Enterprise, the suggested replacements are switched on automatically unless an administrator turned off the global default or disabled a model; no action is needed to remove the retired models.
- This is the third round in a row: 19 October (six models), 16 November (three) and 14 December (six), each announced about a month ahead.
What to do
- Find what you pin. Search your repositories, Copilot agent configurations, custom instructions, scripts and internal docs for the six model names. Anything that selects
Claude Fable 5,Claude Haiku 4.5,Gemini 3.8 Flash,Gemini 4 Argon,Grok 4.6orKimi K3explicitly needs a new value before 14 December. - Pick the replacement. GitHub's table maps Claude Fable 5 to Claude Fable 5.2, Claude Haiku 4.5 to Claude Haiku 5.5, Gemini 3.8 Flash to Gemini 4 Hydrogen, Gemini 4 Argon to Gemini 4 Carbon and Grok 4.6 to Grok 4.7. Kimi K3 has no named successor yet.
- Administrators: if your organisation turned off the global default for new models, the replacements will not appear on their own. Enable them under the model policies in Copilot settings; once enabled they show up in the Copilot Chat model selector. If you explicitly disabled a model, the automatic enablement of its replacement also does not apply.
- Check cost and behaviour before switching. A replacement is a different model, so test the prompts and agent tasks you rely on. Prices and usage multipliers differ by model; GitHub's changelog does not give them, so read the "Models and pricing for GitHub Copilot" page for the current figures.
- Individual annual-plan subscribers who rely on Claude Haiku 4.5 keep it; everyone else should plan to move to Claude Haiku 5.5.
What happened
On 13 November 2026 GitHub published a changelog entry titled "Upcoming deprecation of selected GitHub Copilot models in December". It says six models will be deprecated across all GitHub Copilot experiences on 14 December 2026. The entry names the surfaces: Copilot Chat, inline edits, ask and agent modes, and code completions. It asks users to update their workflows and integrations to supported models before the date.
The table in the entry lists each model with GitHub's suggested alternative:
| Model being retired | Suggested alternative |
|---|---|
| Claude Fable 5 | Claude Fable 5.2 |
| Claude Haiku 4.5 (see exception) | 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 | To be announced |
Claude Haiku 4.5 carries an asterisk. It remains available to individual Copilot subscribers on annual plans, and the deprecation does not apply to those customers. The entry does not give a reason for the exception.
For organisations, the entry repeats the wording GitHub has used in previous rounds. Under default model enablement, the suggested alternatives are enabled automatically for Copilot Enterprise and Business customers. That does not happen if an administrator has turned off the global default or explicitly disabled a model. When the global default is off, admins enable the alternatives through model policies in Copilot settings. No action is needed to remove the retired models themselves, and enterprise customers with questions are pointed to their account team.
Background
The December list is the latest step in a series. GitHub's own changelog entries show the earlier rounds:
- 19 October 2026: Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5 were retired. Suggested alternatives were Gemini 3.8 Flash, GPT-6.1 Sol, GPT-6 Luna and Grok 4.7. GitHub's entry says it first announced these on 18 September 2026.
- 16 November 2026: Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra, announced on 15 October. The suggested alternatives are Claude Opus 5.5 for both Claude models, and GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra. Claude Opus 4.8 carries the same annual-plan exception that Claude Haiku 4.5 carries now.
- 14 December 2026: the six models above, announced on 13 November.
The dates are 28 days apart (19 October to 16 November to 14 December), and each announcement came roughly a month before the retirement: 18 September to 19 October is 31 days, 15 October to 16 November is 32 days, and 13 November to 14 December is 31 days. These are our calculations from the dates in GitHub's entries; GitHub has not said it follows a fixed calendar.
Some of the December models are young. Gemini 3.8 Flash was itself the suggested replacement for Gemini 3.7 Flash in the October round, and Grok 4.7 was the replacement for Grok 4.5. Models named as the destination in one round can be the ones retired in a later round, so "the recommended replacement" is a moving target.
Our own coverage of the new models gives context for what the replacements are: see Claude Fable 5.2 and Mythos 5.2, Gemini 4 Hydrogen, which replaces Gemini 3.8 Flash and Gemini 4 Carbon opening to developers.
The numbers
Counting the models in each of GitHub's three entries shows the size of each round. The October and December rounds are the same size; November was smaller.
| models | |
|---|---|
| 19 Oct | 6 |
| 16 Nov | 3 |
| 14 Dec | 6 |
Source: GitHub changelog entries of 19 October, 15 October and 13 November 2026
Two points from the numbers. First, the vendors affected differ each time: the October round hit OpenAI, Google and xAI models; November hit Anthropic and OpenAI; December hits Anthropic, Google, xAI and Kimi. Second, the list pairs old and new versions of the same lines: Claude Fable 5 goes to Claude Fable 5.2, Claude Haiku 4.5 to Claude Haiku 5.5 and Grok 4.6 to Grok 4.7. GitHub does not say why, but the pattern is that each line keeps its newest versions and older ones are retired.
What it means
For developers who pick a model by name. Hard-coded model names in agent configurations are a liability under a monthly retirement rhythm. Where Copilot offers an automatic selection mode, GitHub's supported-models page says Auto model selection includes Claude Haiku 4.5 and Gemini 3.8 Flash on several surfaces today, so people using Auto are less exposed, though the set it chooses from will change when those models go. Pinned names are the problem.
For administrators. The default-enablement rule is the thing to verify. If your organisation disabled the global default months ago to control which models staff can use, each retirement round now requires a manual step to turn on the new model. Without it, developers will find their usual model gone and no replacement visible in the selector.
For people comparing models. The supported-models page lists Claude Fable 5, 5.1 and 5.2 as generally available, with 1M context and configurable reasoning, and says Kimi K3 also supports 1M context and configurable reasoning. Moving from one to another can change how many AI credits a task uses, because, according to that page, larger context windows and higher reasoning levels use more credits. Budget a trial run before changing a team default.
Who is affected most. Teams using Kimi K3 have the least guidance: GitHub names no replacement. Individual users on annual plans keep Claude Haiku 4.5, which sets those subscribers apart from monthly and organisation customers.
What we don't know yet
- The Kimi K3 replacement. GitHub says "to be announced". We do not know whether it will be a Moonshot model, or whether Copilot will drop the line.
- Why Claude Haiku 4.5 gets an exception. The changelog gives a rule, not a reason. The same exception applied to Claude Opus 4.8 in November.
- Costs of the replacements. The changelog does not list prices or multipliers, and we have not compared them.
- Whether the cadence continues. Three rounds 28 days apart suggest a January round in mid-January, but GitHub has not announced one.
What to watch next
- A further GitHub changelog entry naming a Kimi K3 successor.
- The next deprecation notice, which on the recent pattern would come about a month before a mid-January date.
- Changes to the "supported models" page as Fable 5.2, Haiku 5.5, Gemini 4 Hydrogen, Gemini 4 Carbon and Grok 4.7 become the defaults in Auto selection.
- Whether GitHub extends the individual annual-plan exception to more models.
Sources
- GitHub changelog, 13 November 2026: Upcoming deprecation of selected GitHub Copilot models in December (primary)
- GitHub changelog, 15 October 2026: Upcoming deprecation of selected GitHub Copilot models in mid-November
- GitHub changelog, 19 October 2026: GitHub Copilot models deprecated on October 19, 2026
- GitHub Docs: Supported AI models in GitHub Copilot
Written by
The official account of the people who run audiotale.net: announcements, conversation starters and the newsroom's drafts. Posts written with AI say so. Every other account here is a real person.
Read next
Cloudflare Spend Management beta: hard spending limits for Workers and Durable Objects
Cloudflare's new opt-in Spend Management pauses Workers and Durable Objects when a dollar limit is hit; here is how to set it up and what the pause means.
audiotale.net team5 min read
llama.cpp 1.0.0 and ggml 1.0.0: removed APIs, new server flags, what to change
llama.cpp and ggml reached 1.0.0 on 13 November 2026, removing deprecated functions and adopting strict semantic versioning for the C API.
audiotale.net team6 min read
Gemini Decisions API: gemini-decision-1 scores yes/no and choice questions at $0.10 per million tokens
Google's new endpoint returns calibrated probabilities instead of text, with up to 1,000 questions per request and no output charge.
audiotale.net team6 min read