GitHub will deprecate six Copilot models on October 19

GitHub confirmed, via its official changelog, the deprecation of six Copilot models on October 19, 2026. The change applies to every surface of the tool: Copilot Chat, inline edits, ask and agent modes, and code suggestions in the editor. This isn't a feature change, it's a model retirement: after that date, those names simply disappear from the selector and from API calls, and requests get routed to the replacement indicated by GitHub itself.

What's Leaving and What's Taking Its Place

The published list is straightforward:

| Model | Deprecation Date | Suggested Alternative | |---|---|---| | Gemini 3.7 Flash | 2026-10-19 | Gemini 3.8 Flash | | GPT-5.5 | 2026-10-19 | GPT-5.6 Sol | | GPT-5.4 | 2026-10-19 | GPT-5.6 Sol | | GPT-5.4 mini | 2026-10-19 | GPT-5.6 Luna | | GPT-5 mini | 2026-10-19 | GPT-5.6 Luna | | Grok 4.5 | 2026-10-19 | Grok 4.6 |

It's worth noting a detail that GitHub didn't highlight, but that changes the vocabulary of anyone choosing a model day to day: the two "mini" models (GPT-5.4 mini and GPT-5 mini) converge on a single replacement, GPT-5.6 Luna, while the full versions (GPT-5.5 and GPT-5.4) converge on GPT-5.6 Sol. In practice, the GPT-5.6 family within Copilot seems to have become a pair of profiles (Sol for general use, Luna for light/fast tasks) rather than one version per number, which is consistent with what GitHub had been signaling in another recent changelog, the one about cost and quality configuration in automatic model selection, published days earlier.

How Automatic Migration Works

For anyone using Copilot Business or Enterprise, the switch tends to be invisible. GitHub explains that replacements come enabled by default under the "default model enablement" mechanism: if the organization's administrator hasn't turned off the global default or explicitly blocked that model, the new name automatically appears in the Copilot Chat selector as soon as it becomes available, with no action required. This covers the common case, the developer who picks a model in the interface and doesn't hardcode anything in code.

The problem is for those who turned off the global default on purpose, usually for governance reasons (organizations that want to approve each model manually before releasing it to the team). In these environments, the replacement doesn't appear on its own: someone needs to go into Copilot settings, in the model policies, and explicitly enable GPT-5.6 Sol, GPT-5.6 Luna, Gemini 3.8 Flash, or Grok 4.6. Skipping this step doesn't break Copilot, but it shrinks the list of models available to the team starting October 19, which can catch off guard anyone who only notices when a specific model disappears from the selector mid-session.

Where the Real Pain Shows Up: Model Names Hardcoded in Code

GitHub is explicit on one point: "no action is required to remove the models after they have been deprecated." In other words, the cleanup is automatic on the platform's side. But that only covers those using the standard interface. The real risk is for anyone who referenced the model name outside the UI:

• Extensions or integrations that call the Copilot API passing the model identifier directly (for example, a CI script that invokes gpt-5.4 or grok-4.5 in an automated agent);
• Workspace configurations in VS Code or other editors that pin a specific model in settings.json for a project, a common practice when the team wants reproducibility in agent behavior;
• Internal documentation, runbooks, or versioned prompts that cite the model name as part of the instruction (less critical, but it causes confusion when someone follows the steps and no longer finds the option).

In these cases, the behavior after October 19 isn't well defined by GitHub in the changelog: the most likely outcome, based on how LLM platforms usually handle deprecation, is a silent failure with fallback to the organization's default model, or an explicit "model not found" error on the API call. Since the changelog doesn't detail this point for programmatic use, the safe approach is to treat this as a breaking change: run a grep across the repository and CI configurations for the six names in the table before the date, not after.

Why Swapping Mini for Mini Isn't a Drop-In Replacement

One point that deserves skepticism, not panic: GitHub presents the swap as a direct equivalence ("suggested alternative"), but that's a routing suggestion, not a guarantee of behavioral parity. GPT-5.6 Luna replacing both GPT-5.4 mini and GPT-5 mini suggests consolidation of the light tier, not necessarily the same latency profile, cost per token, or response style that each of the two had on its own. Anyone who chose GPT-5 mini specifically because it was cheaper or faster than GPT-5.4 mini (or vice versa) is losing that distinction; both converge on the same destination.

The same reasoning applies to Gemini 3.7 Flash to Gemini 3.8 Flash and Grok 4.5 to Grok 4.6: these are version increments within the same family, which historically tends to preserve behavior, but "tends to preserve" is not "guarantees." For anyone with automated regression tests over agent output (generated code, automatic review, PR suggestions), it's worth running that suite against the new model before the cutoff, instead of trusting that a higher version label means an identical result.

What the Announcement Leaves Out

The changelog doesn't say what happens to cost: there's no indication that the replacements have the same price per token or the same fit within the Copilot Business/Enterprise budget (in fact, GitHub itself recently launched budget increase requests for Copilot as a separate feature, which suggests that switching models can indeed affect consumption). There's also no deprecation timeline announced for the current models (GPT-5.6 Sol, GPT-5.6 Luna, Gemini 3.8 Flash, Grok 4.6): this quarterly rotation cycle seems to have become routine for Copilot, so the practical checklist item, besides migrating now, is to document internally where each model is referenced in your setup, because this list will repeat itself.

For anyone administering Copilot in an organization with a restrictive policy, the action is clear: review the model policies before October 19. For anyone who just uses the interface, the switch is essentially transparent. For anyone who automated something on top of the model name, this is the deadline to find those references before they break in production.