GitHub Copilot Is Retiring Ten Models by 18 December, and Admins Who Turned Off Default Model Enablement Must Act
Three Copilot models went on 16 November, six go on 14 December and MAI-Code-1.1-Flash on 18 December. GitHub says replacements switch on by default, which leaves admins who turned that default off with work to do.

GitHub has now told Copilot administrators that ten models are leaving the product between Monday 16 November and Friday 18 December 2026. Three went on 16 November, six follow on Monday 14 December and one more, Microsoft’s MAI-Code-1.1-Flash, goes on Friday 18 December. For most organisations the change will pass unnoticed, because GitHub says the suggested replacements switch themselves on. For an administrator who has turned that automatic behaviour off, nothing switches on, and the people who relied on a retired model are left with a picker that no longer offers it.
The notices are spread over three changelog posts that were published on Friday 13 November, Monday 16 November and Tuesday 17 November, so it is easy to read one and miss the others. This piece puts them side by side, explains the one setting that decides whether you have work to do, and says what we would check before 14 December.
#The ten models, in one place
All the dates below are from GitHub’s own changelog. Each retirement applies across Copilot Chat, inline edits, ask and agent modes and code completions, according to GitHub.
| Model | Retired | GitHub’s suggested alternative |
|---|---|---|
| Claude Opus 4.8 | 16 November | Claude Opus 5.5 |
| Claude Opus 5 | 16 November | Claude Opus 5.5 |
| GPT-5.6 Terra | 16 November | GPT-6.1 Sol or GPT-6 Luna |
| Claude Fable 5 | 14 December | Claude Fable 5.2 |
| Claude Haiku 4.5 | 14 December | Claude Haiku 5.5 |
| Gemini 3.8 Flash | 14 December | Gemini 4 Hydrogen |
| Gemini 4 Argon | 14 December | Gemini 4 Carbon |
| Grok 4.6 | 14 December | Grok 4.7 |
| Kimi K3 | 14 December | To be announced |
| MAI-Code-1.1-Flash | 18 December | “Latest MAI Code Flash model” |
Two footnotes matter. First, GitHub says Claude Opus 4.8 and Claude Haiku 4.5 stay available to individual Copilot subscribers on annual plans, so a developer on a personal annual plan may see a model that your organisation’s seats no longer offer. Second, Claude Opus 5 has been removed from the model picker but, GitHub says, remains available for “runtime refusal fallback” through existing authorised access, and existing enterprise model policies continue to apply. We read that as meaning a request can still be handed to Opus 5 behind the scenes if the chosen model declines it, even though nobody can pick it by hand. GitHub’s post does not say how long that lasts.
The 16 November retirements had been announced on 15 October, according to GitHub, so the first three are history. The live deadlines are 14 December and 18 December.
#The setting that decides whether this is your problem
Every one of the three posts repeats the same sentence in different words. Under what GitHub calls default model enablement, the suggested alternatives are switched on automatically for Copilot Enterprise and Copilot Business customers, unless an administrator has turned off the global default or has explicitly disabled the model. If the global default is off, GitHub says, you can enable the alternative models through their model policies in Copilot settings.
That gives three kinds of organisation:
- Default left on, no models disabled by hand. GitHub says the replacements appear in the model selector for your users on their own. Your work is communication: tell developers which model they will be moved to and when.
- Default left on, but one or more models explicitly disabled. A disabled model stays disabled. If your team switched off, say, a Gemini model for data handling reasons and the suggested alternative is from the same vendor, check that the alternative’s policy is what you want, because GitHub’s wording says an explicit disable is respected.
- Global default switched off. This is the careful group, and they are the ones who carry the work. No replacement arrives automatically. Someone must open the model policies in Copilot settings and enable each alternative that your developers will need before the retirement date.
In our view this is the quiet trap in the arrangement. The organisations most likely to have turned the global default off are those that reviewed AI tooling for data protection or procurement reasons, which is exactly the diligence a reader would want to reward. They are also the organisations whose developers will lose a working model on 14 December without an obvious replacement. A control that exists to slow down new models also removes the safety net for old ones.
#MAI Code 2 arrived on the same terms
The 17 November post announcing MAI Code 2 and MAI Code 2 Flash in GitHub Copilot follows the same pattern. According to GitHub, MAI Code 2 is available on Copilot Pro, Pro+, Max, Business and Enterprise, and MAI Code 2 Flash on all of those plus Free and Student. Business and Enterprise administrators manage access through the model policy in Copilot settings, and under default model enablement the new models are enabled automatically unless the administrator has turned off the global default or explicitly disables the model.
That means that, for an organisation on the default, two things can have happened on 17 November without anyone being asked: two new models became available to developers, and a model that developers may have chosen, MAI-Code-1.1-Flash, was given an end date of 18 December. GitHub says Copilot Enterprise administrators “may need” to enable the replacement through their model policies.
Cost is the part that deserves attention. GitHub says both models are billed under usage-based billing. InfoWorld, citing GitHub’s pricing page, reports promotional prices for MAI Code 2 of $0.50 per million input tokens, $0.025 per million cached input tokens and $2.50 per million output tokens until 17 May 2027, rising to $1.00, $0.05 and $5.00 afterwards. For MAI Code 2 Flash it reports $0.05, $0.005 and $0.25 per million tokens, rising to $0.10, $0.01 and $0.50. A token is the small chunk of text, roughly a short word or part of one, that a model reads and writes, and usage-based billing means the bill follows how many of them your developers use.
The comparisons Microsoft published come with caveats, and InfoWorld reports them. Microsoft said MAI Code 2 averaged a 63.2% pass rate across four coding benchmarks, against 64.1% for OpenAI’s GPT-6.1 Sol and 65.3% for Anthropic’s Claude Opus 5.5. These are Microsoft’s figures, and the cost-per-task comparison InfoWorld quotes was measured at the 50% promotional price. Greyhound Research’s Sanchit Vir Gogia told InfoWorld that the gap narrows sharply once the promotion ends, and that leaders should test a model on their own code, through their own tools, before adopting it. We agree with that, and would add that a model enabled automatically by default skips the testing step unless you create one.
#What we would check before 14 December
- Find out which setting you have. In Copilot settings, look at the model policies and the global default for your organisation or enterprise. Write down whether it is on or off, and which models are explicitly disabled. That single fact tells you whether you are in the group that has to act.
- Find who is actually using the retiring models. GitHub’s post says no action is needed to remove a model once it is deprecated, so the risk is not a failed removal. It is a developer, an agent configuration or a shared prompt that names a retired model and quietly falls back to something else. Search repositories and team documents for the model names in the table, particularly Claude Fable 5, Gemini 3.8 Flash and Grok 4.6, which share the 14 December date.
- Enable the replacements you want, and only those. If your global default is off, turn on the alternatives in the table that you are content to pay for and allow. Kimi K3 has no named replacement yet; GitHub lists it as “to be announced”, so decide whether your developers need to be told to pick another model themselves.
- Watch the bill after the switch. Because GitHub says the MAI models are billed under usage-based billing, compare spend in the first week of December with the week before. The promotional prices are time limited, as InfoWorld reports.
- Put the three dates in a calendar. 14 December and 18 December are the retirement dates, and 17 May 2027 is when InfoWorld reports the MAI promotional prices end. On this evidence models are being retired roughly monthly, so a standing review is cheaper than a scramble each time.
#If you buy Copilot Business for a small team
A firm with a handful of developers on Copilot Business is covered by the same rules as an enterprise, according to GitHub: the retirements apply across Copilot, and the model policy in Copilot settings is where access is managed. Nobody needs a specialist. The person who administers the account should look once at the global default, and if it has never been touched, GitHub says the replacements will arrive on their own. If it was turned off, they have to enable the replacements by hand before 14 December.
#Our view: a faster treadmill, and a default that cuts both ways
We think the real story is the pace rather than any single retirement. Ten models retired in under five weeks, each with roughly a month’s notice from GitHub’s announcement, is a rhythm that most change-managed IT teams are not built for. It sits uneasily beside governance work that takes quarters. Our expectation is that GitHub will keep retiring models on a similar cadence, because the catalogue it offers keeps growing, and that administrators will increasingly want a named owner and a monthly review for model policy, just as they have one for licences.
We also think default enablement is a sensible choice for GitHub and an uncomfortable one for cautious customers. It keeps developers working. But it moves the decision about which new model a developer may use from a procurement conversation to a changelog post. We would rather see the opposite default for regulated teams: a new model off until someone approves it, with the retirement notice naming each person who must act. That is not how GitHub describes it, and what we have set out is our reading of the changelog, not a statement from GitHub. We would be proved wrong if the retirements slowed to a few a quarter.
For readers who have followed this site’s coverage of defaults in Microsoft’s Copilot, the pattern will be familiar. We covered Microsoft’s decision to switch Claude on by default in Microsoft 365 Copilot from 9 December, where the same tension between convenience and control is at work. In both cases the practical advice is the same: know what your default is before the vendor’s date arrives.
#What we do not know
GitHub’s posts do not say how many organisations have the global default switched off, how long Claude Opus 5 will remain available for refusal fallback, or what will replace Kimi K3. They do not give prices for the replacements named in the table other than the two MAI models covered by InfoWorld. We have not tested any of the models and make no claim about their quality beyond the vendor figures we have attributed.
Evidence trail
Sources and verification
Primary documentation and external technical references used in this article.
- 01GitHub changelog, 13 November 2026: Upcoming deprecation of selected GitHub Copilot models in Decembergithub.blog
- 02GitHub changelog, 16 November 2026: GitHub Copilot models deprecated on November 16, 2026github.blog
- 03GitHub changelog, 17 November 2026: Upcoming deprecation of MAI-Code-1.1-Flash in Decembergithub.blog
- 04GitHub changelog, 17 November 2026: MAI Code 2 and MAI Code 2 Flash in GitHub Copilotgithub.blog
- 05InfoWorld: Microsoft launches MAI Code 2 models to challenge OpenAI and Anthropic in AI codinginfoworld.com
About this page
Credited to
Dates
Published
Corrections
No corrections are listed for this page. Some review sweeps corrected pages without naming each one; those are in the corrections log.
More news
All newsWas this useful?
Reports go to the site owner and are used to correct pages. Corrections are listed at /corrections.
Get new fixes by email.
New runbooks, configuration traps and fixes from KBY Technologies, about twice a week.




Comments
Add a thoughtful note on GitHub Copilot Is Retiring Ten Models by 18 December, and Admins Who Turned Off Default Model Enablement Must Act. Comments are checked for spam and held for moderation before appearing.