GitHub Copilot Retires Six Models on 14 December: What Each Swap Costs
On 13 November 2026 GitHub announced that six models will be deprecated "across all GitHub Copilot experiences" on 14 December 2026: chat, the coding agent, the CLI, the app and the IDE integrations. Each has a suggested replacement except Kimi K3, whose replacement is "to be announced". The same day GitHub made Gemini 4 Hydrogen generally available in Copilot, which is the named successor to Gemini 3.8 Flash.
Model churn in Copilot is not new. What is new is that Copilot now bills by token through AI credits (one credit is $0.01, per GitHub's models and pricing page), so a forced model swap is also a price change. Some of these swaps cut your bill by 90 percent. One doubles it.
What happened
| Deprecated on 14 Dec | Suggested replacement | Input / cached / output per 1M tokens, before | After |
|---|---|---|---|
| Claude Fable 5 | Claude Fable 5.2 | $10 / $1 / $50 | $8 / $0.20 / $40 |
| Claude Haiku 4.5 | Claude Haiku 5.5 | $1 / $0.10 / $5 | $0.10 / $0.01 / $0.50 for prompts up to 100K tokens; $0.50 / $0.05 / $2.50 above |
| Gemini 3.8 Flash | Gemini 4 Hydrogen | $0.75 / $0.075 / $3.75 | $1.50 / $0.075 / $7.50 |
| Gemini 4 Argon | Gemini 4 Carbon | $2 / $0.10 / $10 | Same until 1 Dec 2026, then $4 / $0.20 / $20 |
| Grok 4.6 | Grok 4.7 | $2 / $0.50 / $6 | Same for prompts up to 200K tokens |
| Kimi K3 | To be announced | Not checked | None named |
Prices are from GitHub's models and pricing page as read on 16 November. They match the providers' own list prices on my LLM API pricing tracker, which is what the Gemini 4 Hydrogen changelog says Copilot charges: "provider list pricing under usage-based billing."
There is one exception. GitHub says Claude Haiku 4.5 "will remain available to individual GitHub Copilot subscribers on annual plans." Everyone else loses it on 14 December.
What changes for engineers
Cost, by swap. Take a fixed workload of one million input tokens and 100,000 output tokens, none of it cached, and price it before and after:
| Swap | Before | After | Change |
|---|---|---|---|
| Fable 5 to Fable 5.2 | $15.00 | $12.00 | 20% cheaper |
| Haiku 4.5 to Haiku 5.5 (prompts up to 100K) | $1.50 | $0.15 | 90% cheaper |
| Gemini 3.8 Flash to Gemini 4 Hydrogen | $1.125 | $2.25 | Twice the price |
| Gemini 4 Argon to Gemini 4 Carbon, after 1 Dec | $3.00 | $6.00 | Twice the price |
| Grok 4.6 to Grok 4.7 | $2.60 | $2.60 | No change |
The Argon line needs a footnote. Google is doubling Argon's own API price on 1 December too, as I covered in the Gemini 4 price post, so the Carbon swap costs nothing extra compared with staying on Argon for its last two weeks. The Flash swap is the one that actually raises the bill, two weeks before Google's own doubling of gemini-3.8-flash on the API on 1 January 2027 would have done the same. Teams that picked Gemini 3.8 Flash as the cheap model for high-volume agent work will see that line double unless they move to Haiku 5.5, which at $0.10 input for short prompts is now cheaper than Flash ever was.
These are list-price calculations on a made-up workload, not measurements. Your real cost depends on caching, prompt length and how many turns the agent takes, and newer models do not always use the same number of tokens for the same task.
Defaults move for you, unless you turned defaults off. GitHub says the replacements are "automatically enabled for Copilot Enterprise and Copilot Business customers unless an administrator has turned off the global default or explicitly disabled the model." If your organisation curates the model list by hand, which many do after a security review, the replacement will not appear and users will simply lose the model on 14 December. Nothing needs to be done to remove the old models.
Fable 5.2 carries a data-retention note. GitHub's supported models page footnotes the Fable models: zero data retention is available only under a time-limited exemption to the end of 2026, and an administrator must enable each model before users can use it. If your security sign-off for Fable 5 depended on zero retention, re-check it for 5.2 before you enable it.
Pinned model names will break quietly. Neither changelog mentions custom agents, prompt files or CLI configuration that name a model. Anything in your repositories that pins one of the six by name is worth finding before 14 December rather than after.
What I ran
Nothing. There is no release to install. The price comparisons above are arithmetic on GitHub's published per-token prices.
What to do this week
- Check the model policy page in your Copilot organisation or enterprise settings. If the global default is off, enable Claude Fable 5.2, Claude Haiku 5.5, Gemini 4 Hydrogen, Gemini 4 Carbon and Grok 4.7 deliberately, or decide not to.
- Search for pinned names. Run
git grep -nE "Haiku 4\.5|Gemini 3\.8 Flash|Gemini 4 Argon|Grok 4\.6|Kimi K3|Fable 5([^.]|$)"across repositories that hold agent definitions, prompt files and CI workflows that call Copilot. - Re-price your cheap tier. If Gemini 3.8 Flash was your default for bulk agent work, compare Haiku 5.5 and Hydrogen on a sample of your own tasks before 14 December. The LLM cost calculator reads the same list prices.
- Plan for Kimi K3 users. There is no replacement yet. If anyone relies on it, pick a fallback now.
- Set a spend alert on AI credits before the swap, so a doubled Flash bill shows up in a week, not at month end.
FAQ
Which models is GitHub Copilot deprecating in December 2026?
Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3, all on 14 December 2026. The suggested replacements are Claude Fable 5.2, Claude Haiku 5.5, Gemini 4 Hydrogen, Gemini 4 Carbon and Grok 4.7. Kimi K3's replacement has not been announced.
Will Claude Haiku 4.5 stay in Copilot for anyone?
Yes. GitHub says Claude Haiku 4.5 remains available to individual Copilot subscribers on annual plans, and the deprecation does not apply to them.
Is Gemini 4 Hydrogen more expensive than Gemini 3.8 Flash in Copilot?
Yes. GitHub lists Gemini 3.8 Flash at $0.75 input and $3.75 output per million tokens and Gemini 4 Hydrogen at $1.50 and $7.50, so the same token count costs twice as much. Cached input is the same at $0.075.
Related: GitHub Universe 2026: Copilot Fusion Launches, AI Model Release Tracker and LLM API Pricing Tracker.
Spot an error or something out of date? Email me and I will fix it and log it on the corrections page.
I'm Ajay Kumar — I build and operate PandaStack, an open-source Firecracker microVM cloud for AI agents, and write about the infrastructure and tooling around it.
Need this kind of infrastructure work? See what I do or email hello@ajayk.sh.
Related
OpenAI Models Leave Cursor on 12 November as Cursor Ships Agent SDK
OpenAI's proposed cutoff of its models in Cursor lands on 12 November 2026, after SpaceX bought Cursor maker Anysphere. Meanwhile Cursor shipped an Agent SDK, Environments API and live apps. What teams should do.
6 minNov 10, 2026Uber Cuts 3,300 Jobs, About 10% of Staff, and Its Memo Never Mentions AI
On 2 September 2026 Uber said it would cut about 10 percent of its workforce, roughly 3,300 roles, mostly management layers, weeks after a $2.4 billion quarterly profit. The memo skips AI; the context around it does not.
6 minNov 10, 2026Alphabet Q3 2026: $51.4B Capex, Cloud Up 103%, Free Cash Flow at -$9.1B
Alphabet's Q3 2026 results: $51.4 billion of quarterly capex, Google Cloud revenue up 103% to $30.7 billion, negative free cash flow of $9.1 billion, an $802 billion commitments book and new debt and equity to pay for it.
7 min