GitHub Copilot retires 19 models across four waves, three more exit on November 16

Nineteen models, four waves, ten weeks

GitHub Copilot is retiring 19 AI models across four waves in roughly ten weeks, according to a dev.to article whose author says every row was read directly from GitHub's changelog posts. Ten models were removed on October 2 and October 19, three more disappear on November 16, and a final six go on December 14. The author's warning is blunt: if you deliberately picked a Copilot model at any point this year, there is a real chance it is on the list.

Waves one and two: ten models already gone

The October 2 wave removed four models. Gemini 3.5 Flash and Gemini 3.6 Flash were both replaced by Gemini 3.8 Flash, Kimi K2.7 Code was replaced by Kimi K3, and Claude Opus 4.7 was replaced by Claude Opus 5.5.

The October 19 wave removed six: Gemini 3.7 Flash (to Gemini 3.8 Flash), GPT-5.5 and GPT-5.4 (to GPT-6.1 Sol), GPT-5.4 mini and GPT-5 mini (to GPT-6 Luna), and Grok 4.5 (to Grok 4.7). The post highlights a discrepancy worth knowing about: GitHub's original September 18 announcement named GPT-5.6 Sol, GPT-5.6 Luna and Grok 4.6 as the replacements, but by retirement day the changelog listed GPT-6.1 Sol, GPT-6 Luna and Grok 4.7. Some third-party summaries still quote the outdated September names.

Waves three and four: what is still coming

On November 16, Claude Opus 4.8 and Claude Opus 5 are replaced by Claude Opus 5.5, and GPT-5.6 Terra is replaced by GPT-6.1 Sol or GPT-6 Luna. One exception applies: Claude Opus 4.8 stays available to individual Copilot subscribers on annual plans, so that group keeps an Opus option.

December 14 retires six more. Claude Fable 5 gives way to Claude Fable 5.2, and Claude Haiku 4.5 to Claude Haiku 5.5, with Haiku 4.5 excepted for individual annual subscribers in the same way. Gemini 3.8 Flash moves 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 at all; the changelog entry reportedly reads only "to be announced".

The replacements are churning too

The article's sharpest finding is that three of the named successors are themselves on the retirement list. Gemini 3.8 Flash, the destination for three October retirements, disappears on December 14 in favour of Gemini 4 Hydrogen. Kimi K3 and Grok 4.6, both cited as replacements in earlier waves, also go on December 14, and Kimi K3's successor remains unnamed. Half of the December wave therefore consists of models GitHub itself directed users toward within the previous two months, which the author frames as a roughly two-month shelf life for any "safe" model choice.

What breaks on the day

According to the post, deprecated models simply stop appearing wherever Copilot exposes a picker, including chat, agent mode and inline edits, and GitHub states no action is required to remove them after deprecation. Migration on the user's side is not automatic. Configs, integrations and pinned settings that name a retired model will either fail or fall back to a default, depending on how they were written. For Copilot Enterprise and Copilot Business customers, suggested replacements are enabled automatically under default model enablement, unless an administrator turned off the global default or explicitly disabled a specific model; in that case the replacements must be re-enabled through model policies in Copilot settings before the deadline.

What to do before each deadline

Users of Claude Opus 5, Opus 4.8 or GPT-5.6 Terra should test the replacements immediately, since the successors are already available and may behave differently on their kind of work. Kimi K3 users are the worst positioned: with nothing to switch to yet, the author recommends evaluating the other models now rather than choosing blind on deadline day. Gemini 4 Hydrogen launched on November 12 with a one-million-token context window and listed pricing of $1.50 per million input tokens and $7.50 per million output tokens, positioning it as the fast-and-cheap successor to the Flash line. Team administrators should search repositories and configuration files for the nineteen model names and update them, the step the author says most people skip and the one that actually breaks. Users who never chose a model and simply typed into Copilot Chat are likely fine, since defaults are updated for them.

Why it matters

This is not a one-off cleanup but evidence of a sustained retirement cadence, and it changes how teams should configure AI tooling. Hardcoding a model name into a script, pipeline or editor setting is now a bet with a two-month expiry, and three of GitHub's own recommended destinations have already been invalidated by later waves. Teams on Enterprise and Business plans with restrictive model policies face silent breakage if they do not re-enable replacements in time. The practical takeaway from the post: point tooling at a role, such as a fast or strong model, rather than a specific name, and calendar the Copilot changelog so the next wave does not arrive as a surprise.