GitHub Copilot Drops Claude Opus 4.8, Opus 5, GPT-5.6 Terra Today

GitHub Copilot deprecated Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra today, Nov 16 2026. Here's who's affected, the replacements, and what to switch before workflows break.

GitHub Copilot illustration accompanying GitHub's announcement of AI model changes
Illustration from GitHub's Copilot model changelog. Image: GitHub.

GitHub Copilot removed three AI models from every Copilot surface today, November 16, 2026: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra. GitHub posted the deprecation notice a month in advance, on October 15, and the cut now applies across Copilot Chat, inline code completions, ask mode, and agent mode. If your workflow, extension, or CI pipeline is still pinned to one of those three model IDs, it stops resolving today — switch to the listed replacements now.

What happened
  • Deprecated today (Nov 16, 2026): Claude Opus 4.8, Claude Opus 5, GPT-5.6 Terra
  • Replacements: Claude Opus 5.5 (for both Opus models); GPT-6.1 Sol or GPT-6 Luna (for GPT-5.6 Terra)
  • Exception: individual Copilot subscribers on annual plans keep Claude Opus 4.8
  • Scope: Copilot Chat, inline edits, ask/agent mode, and code completion, on every plan
  • Next wave already announced: six more models are scheduled to deprecate on Dec 14, 2026

What GitHub announced

The notice ran under GitHub's standard changelog format for model lifecycle changes: a short post titled "Upcoming deprecation of selected GitHub Copilot models in mid-November," published October 15, 2026, giving customers roughly a month's notice. It names three models and gives all of them the same deprecation date, November 16, 2026 — today. GitHub's language is consistent with how it has handled previous model retirements: the models don't get pulled quietly, they get a dated changelog entry, a suggested-replacement table, and an explicit statement that no cleanup action is required afterward, since Copilot simply stops offering the retired model in the picker and for API-style integrations.

The three models affected:

  • Claude Opus 4.8 — Anthropic's earlier Opus-tier model, long available in Copilot's model picker for Chat and agent workflows.
  • Claude Opus 5 — the direct predecessor to Opus 5.5, also an Opus-tier reasoning model.
  • GPT-5.6 Terra — one of the GPT-5.6 family variants GitHub had offered alongside Luna and Sol.

The annual-plan exception for Claude Opus 4.8

There's one carve-out worth flagging separately, because it's easy to miss in a skim of the changelog: individual Copilot subscribers on annual billing plans keep access to Claude Opus 4.8. GitHub's own footnote on the deprecation table says the removal "doesn't apply" to that specific group. If you're on an individual annual plan, Opus 4.8 isn't going anywhere for you today — but Claude Opus 5 and GPT-5.6 Terra still are, and so is Opus 4.8 for anyone on a monthly individual plan, a Business seat, or an Enterprise seat.

That asymmetry matters for teams: a developer on a personal annual Pro or Pro+ subscription may still see Opus 4.8 in their model picker today while a colleague on the same org's Business seats does not. Don't assume consistent behavior across a team just because everyone uses the same editor.

What to switch to

GitHub's changelog maps each deprecated model to a specific suggested alternative, and the current status of every model GitHub supports in Copilot — including the three replacements — is tracked on GitHub's supported AI models reference page. The table below is the full deprecation mapping as published.

Deprecated modelDeprecation dateSuggested alternativeNotes
Claude Opus 4.8Nov 16, 2026Claude Opus 5.5Stays available to individual subscribers on annual plans
Claude Opus 5Nov 16, 2026Claude Opus 5.5No exception
GPT-5.6 TerraNov 16, 2026GPT-6.1 Sol or GPT-6 LunaEither replacement is listed; no single default is specified

For the two Claude Opus models, there's a single recommended target: Claude Opus 5.5. If your prompts, system instructions, or evaluation harnesses were tuned against Opus 4.8 or Opus 5 specifically, budget time to re-test against 5.5 rather than assuming a drop-in swap — reasoning-tier model upgrades can shift output length, tool-call behavior, and refusal patterns even when the vendor frames it as a direct successor.

For GPT-5.6 Terra, GitHub gives two acceptable replacements rather than one: GPT-6.1 Sol or GPT-6 Luna. That's useful if you want to A/B the two before standardizing, but it also means there's no single "correct" migration target — pick based on whichever your team already benchmarks better for your workload (long-context agent tasks vs. fast inline completions tend to favor different members of a model family).

Who's actually affected right now

This lands differently depending on plan type:

  • Free and individual Pro/Pro+ users on monthly billing: all three models disappear from the picker today. Any saved chat session, custom instruction, or extension config that hardcodes one of the three model names needs updating.
  • Individual users on annual plans: Claude Opus 4.8 keeps working; Claude Opus 5 and GPT-5.6 Terra still deprecate today.
  • Business and Enterprise customers: GitHub says the suggested alternatives — Opus 5.5, Sol, and Luna — are enabled by default for these plans, unless an admin has already turned off the global default for new models or explicitly disabled one of them. If your org has locked down model policies, an admin needs to check the Copilot policy settings and turn the replacements on manually before developers can use them.

In every case, GitHub is explicit that no action is required to remove the deprecated models themselves — they simply stop being selectable or stop resolving in API-style calls after the cutoff. The only action item is on the adoption side: move any pinned configuration to a supported model before (ideally; now, since the date has passed) workflows start failing.

GitHub's own phrasing on this is direct: users should move any workflows or integrations built against the deprecated models to a supported one before the deprecation date, and "no action is needed to remove the models" once that date passes — GitHub handles the removal on its end. In practice, that means the burden sits entirely on finding and updating anything on your side that names the old model, not on cleaning anything up after the fact.

Why GitHub keeps retiring Copilot models

This isn't a one-off. GitHub has been running a rolling model-lifecycle cadence in Copilot for a while now, adding new frontier models as vendors ship them and retiring older ones roughly a cycle or two later. In the same stretch of November, GitHub also shipped new models into Copilot — Claude Fable 5.2 landed November 9, Gemini 4 Carbon on November 10, and Gemini 4 Hydrogen on November 13 — while simultaneously winding down the older Opus and GPT-5.6 variants being retired today.

GitHub has also already posted the next wave: a changelog entry published November 13 names six more models scheduled to deprecate on December 14, 2026 — Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6, and Kimi K3 — with Haiku 4.5 carrying the same "stays available to annual individual subscribers" exception as Opus 4.8 does today, and the suggested replacement for Kimi K3 listed as "to be announced." Teams that just finished migrating off today's three models should treat that December date as the next checkpoint rather than a separate, unrelated event.

The practical takeaway: pinning a specific Copilot model ID in a script, extension, or CI job is inherently a short-lived choice. If your team runs Copilot-dependent automation, build the model name as a configurable value you can swap in minutes, not something buried in a dozen files.

How to switch your model today

For Copilot Chat in your IDE, the model picker sits in the chat panel itself — GitHub's own how-to for changing the Copilot Chat model walks through where that control lives across VS Code, Visual Studio, and JetBrains IDEs. Inline code completion has a separate model setting, documented in GitHub's guide to changing the completion model. If you administer Copilot for a Business or Enterprise org, the place to re-enable or restrict specific models is the Copilot model policy page in your organization or enterprise settings — that's also where you'd check whether the global "enable new models by default" toggle is on, since that setting determines whether Opus 5.5, Sol, and Luna were already switched on for your users today without anyone touching anything.

It's worth doing this audit alongside other recent Copilot-adjacent platform changes rather than in isolation — GitHub has shipped a steady stream of these this quarter, from GitHub's Slack and Teams integration going generally available to policy-level changes like the one that pauses Dependabot alerts on inactive repositories starting December 8. None of those are related to model deprecation directly, but they're part of the same pattern: GitHub is iterating on Copilot and its surrounding platform fast enough that a monthly settings review is no longer optional for teams that depend on it.

Does switching models change your usage allowance?

One detail worth checking before you standardize on a replacement: Copilot doesn't treat every model as consuming the same amount of your plan's allowance. GitHub's billing documentation describes a per-model "premium request multiplier" — each model is assigned a multiplier based on its complexity and resource cost, and that multiplier is applied against your monthly allotment of premium requests or AI credits every time you use it. A heavier reasoning model can burn through a Pro or Business allowance faster than a lighter one, even for comparable prompts.

GitHub's plan tiers themselves haven't changed as part of this deprecation: Copilot Free remains limited to auto-selected models, Pro is $10 per user per month, Pro+ is $39 per user per month with access to premium models, Business is $19 per seat per month, and Enterprise is $39 per seat per month. What can change is how far your existing allowance stretches once your team's traffic shifts from Opus 4.8/Opus 5 to Opus 5.5, or from GPT-5.6 Terra to Sol or Luna, since the new models may carry different multipliers than the ones they replace. GitHub's plan comparison page lays out the credit allotments per tier; if your org tracks Copilot spend closely, it's worth checking your usage dashboard a week or two after today's cutover rather than assuming costs stay flat.

Editor and tool-specific wrinkles

If your team compares Copilot's model options against other in-editor assistants — for instance, in the kind of side-by-side testing covered in our look at VS Code 1.146's Copilot Fusion comparisons — keep in mind that benchmark numbers gathered against Opus 4.8, Opus 5, or GPT-5.6 Terra are now measuring a retired baseline for most users. Re-run any internal comparison that cites those three models by name; the results may no longer reflect what's actually selectable in your editor.

Extensions and third-party tools that call into Copilot's model list programmatically should also be checked. If a tool hardcodes a model identifier rather than reading from Copilot's live picker, it will either silently fall back to a default model or throw an error today, depending on how it was built. This is the kind of failure that doesn't show up until someone actually runs the affected command, so it's worth a quick grep across your team's tooling for the three retired model names.

What to do next

Three concrete steps, in order of urgency:

  1. Check what's pinned. Search your repos, CI configs, and Copilot extension settings for "Opus 4.8," "Opus 5," and "GPT-5.6 Terra." Anything hardcoded to those strings needs updating now, not on a future sprint.
  2. Confirm your org's policy state. If you're a Copilot admin on a Business or Enterprise plan, open your model policy settings and confirm Opus 5.5, GPT-6.1 Sol, and GPT-6 Luna are actually enabled — don't assume the "enabled by default" behavior applies if you've previously customized model policies.
  3. Mark your calendar for December 14. The next deprecation wave — Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6, and Kimi K3 — is already scheduled. Treat today's migration as a template for that one.

Frequently asked questions

Which GitHub Copilot models were deprecated on November 16, 2026?

Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra, across Copilot Chat, inline completions, ask mode, and agent mode.

What should I switch to instead of Claude Opus 4.8 or Claude Opus 5?

GitHub's changelog recommends Claude Opus 5.5 as the replacement for both deprecated Opus models.

What's the replacement for GPT-5.6 Terra in Copilot?

GitHub lists two acceptable alternatives: GPT-6.1 Sol or GPT-6 Luna. No single default is specified, so teams can pick based on their workload.

Does anyone still have access to Claude Opus 4.8?

Yes. Individual Copilot subscribers on annual billing plans keep access to Claude Opus 4.8; the deprecation doesn't apply to that group. Claude Opus 5 and GPT-5.6 Terra deprecate for everyone, including annual individual subscribers.

Do I need to do anything to remove the deprecated models myself?

No. GitHub says no action is needed to remove the models after the deprecation date — they stop being selectable automatically. The only required action is updating any workflow or integration still pointed at the old model names.

Are more Copilot model deprecations coming after this one?

Yes. GitHub has already announced a second wave for December 14, 2026, deprecating Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6, and Kimi K3.

Sources

More on GitHub Copilot →GitHub CopilotClaude OpusGPT-5.6 TerraAI model deprecationDeveloper tools
Felix Moreau
Written byFelix Moreau

Felix Moreau writes Pandromeda's software coverage and how-to guides. He covers Windows, macOS and Linux updates, the apps people rely on, emulators and developer tools, and turns official documentation into clear, numbered steps that work on the current version.

More from Software & Guides

See all