GitHub Copilot Drops 3 AI Models Nov 16, 2026: Migrate Now

On November 16, 2026, three AI models leave GitHub Copilot for good: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra. If your workflow, extension settings, or CI job names any of those three models directly, it stops working as written on that date — or silently falls back to whatever GitHub assigns in its place. GitHub announced the retirement on October 15, 2026, and gave developers about a month to move.

Here’s the short version: switch Opus 4.8 and Opus 5 to Claude Opus 5.5, and switch GPT-5.6 Terra to GPT-6.1 Sol or GPT-6 Luna. There is exactly one exception — individual subscribers on annual plans keep Opus 4.8. Everyone else loses all three. The rest of this article is the migration checklist: what to search for, what to retest, and the admin settings that decide whether your team even sees the replacements.

I’ve watched Copilot’s model menu change all year, and this is the third retirement wave in about two months. The pattern is now routine: a new model joins, overlaps with the old one for a while, then the old one gets a fixed sunset date. What makes this round sting is that it removes two Opus models at once, and Opus is the model a lot of developers reach for when a task is genuinely hard.

What exactly leaves Copilot on November 16

GitHub’s October 15 changelog entry is precise about scope. The three models leave “all GitHub Copilot experiences,” which GitHub defines as Copilot Chat, inline edits, ask and agent modes, and code completions. That matters, because an earlier wave of retirements mostly affected the chat picker. This one reaches everywhere Copilot runs — including the agent mode workflows and custom agents that teams have built up over the year.

Here is the official mapping:

Leaving on November 16, 2026	GitHub’s suggested replacement
Claude Opus 4.8*	Claude Opus 5.5
Claude Opus 5	Claude Opus 5.5
GPT-5.6 Terra	GPT-6.1 Sol or GPT-6 Luna

The asterisk on Opus 4.8 is the exception, and we’ll get to it in a moment. First, notice what the table does not say: it does not promise the replacement behaves identically. Opus 5.5 is a newer model, not a renamed Opus 5. GPT-6.1 Sol and GPT-6 Luna are different models from GPT-5.6 Terra, with different strengths. GitHub is telling you where to go, not guaranteeing the trip feels the same. That is why retesting — not just renaming — is the core of this migration.

One more thing the changelog makes clear: “No action is required to remove the models after they have been deprecated.” GitHub pulls them from the picker itself. Your job is not deletion. Your job is making sure something working is waiting in their place.

The one exception: annual individual subscribers keep Opus 4.8

In GitHub’s words, “Claude Opus 4.8 will remain available to individual GitHub Copilot subscribers on annual plans so they retain an Opus offering. The deprecation of Claude Opus 4.8 does not apply to these customers.”

Read that carefully, because three common misreadings are already circulating:

1. It covers Opus 4.8 only. Nobody keeps Opus 5 or GPT-5.6 Terra under this exception — not annual subscribers, not anyone.
2. It covers individuals only. If your seat is billed through Copilot Business or Copilot Enterprise, you lose all three models on November 16, even if you personally pay annually for something else. Organization-managed seats are not covered.
3. It covers annual plans only. Monthly individual subscribers lose Opus 4.8 like everyone else.

There is also a billing wrinkle worth knowing if you are on this path. For annual Pro and Pro+ subscribers still on request-based billing, GitHub’s annual-plan multiplier table lists Opus 4.8 at a multiplier of 27 — meaning each request counts as 27 premium requests. Keeping Opus 4.8 is not free in budget terms; it is expensive in request terms. If you stay on Opus 4.8 because a workflow depends on it, watch your premium-request usage after November 16 the way you’d watch a meter running.

My honest take: treat the exception as a grace period, not a permanent home. GitHub kept Opus 4.8 for annual subscribers to avoid breaking long-running individual workflows mid-subscription. It has not promised to keep it forever, and every previous Opus model eventually left Copilot entirely. If you can move to Opus 5.5 now, do it — and keep Opus 4.8 only for the one or two workflows where the new model genuinely behaves worse.

What happens if you do nothing

This is the question every developer actually wants answered, so let’s be blunt about the three scenarios:

If you pick models from the Copilot Chat picker by hand: you’ll be fine, mostly. The retired models disappear from the menu, and Copilot may quietly move your chat to Auto or a default model. You’ll notice the change, you might grumble about it, and you’ll pick a new model. No breakage, just friction.

If you use agent mode, custom agents, or saved prompts with a pinned model: this is where breakage lives. Any workflow that explicitly calls Claude Opus 5, Claude Opus 4.8, or GPT-5.6 Terra by name will fail or silently fall back to a default after November 16. “Silently fall back” is the worse outcome, because your agent keeps running — just on a model you didn’t choose, with behavior you didn’t test. A code-review agent that was tuned on Opus 5’s judgment, now running on a different model, can produce subtly different reviews and nobody notices for weeks.

If you run Copilot in CI or scripted integrations: same risk, higher stakes. A script that passes a model name to a Copilot API or CLI call can start erroring on November 16, and if that script gates a deploy or a nightly job, you find out at the worst possible time.

The fix for all three scenarios is the same: find every place the old model names appear, switch them, and retest the workflow on the replacement before the deadline. The sections below walk through each audience.

Step 1: Find every place the old model names appear

Before you change anything, search. The old model names hide in more places than most people expect:

• Editor settings: VS Code settings.json, JetBrains IDE settings, and any .vscode folder settings that pin a Copilot model for chat or completions.
• Custom agents and skills: agent definition files, prompt files, and instruction files that name a model. If your team built custom Copilot agents this year, check each one’s configuration.
• Prompt libraries: shared prompt packs, internal wikis, and README files that tell teammates “use Opus 5 for this task.”
• CI and automation: GitHub Actions workflows, scripts, and scheduled jobs that call Copilot features with a model parameter.
• Dotfiles and templates: starter repos, team templates, and onboarding docs that bake in a model choice for new hires.
• Documentation: runbooks that say “ask Copilot (Opus 5) to…” — these won’t break anything technically, but they’ll confuse the next person who follows them.

A practical search: run a text search across your repos for the strings Opus 4.8, Opus 5, GPT-5.6, and Terra. Watch out for false positives — “Opus 5” can appear in unrelated contexts — but a case-insensitive grep across your working repos takes minutes and catches the real references. Do the same for your dotfiles repo and your team’s shared prompt library. If you manage a team, ask each developer to search their own local config too, since editor settings often live outside version control.

One caution from the retirement history: GitHub has retired models before, and Copilot has sometimes moved chats to Auto when the old model went away. Don’t assume your picker state survived a previous wave untouched. Open the picker and look.

Step 2: Switch to the replacements — and pick the right Terra successor

For the Opus models, the choice is simple: Opus 5.5 replaces both. GitHub positions Opus 5.5 as the current Opus generation in Copilot, and it is available on Pro+ and higher tiers (Copilot Pro at $10 a month does not include Opus models at all — on Pro you get Claude Haiku and Sonnet models, so if you never saw Opus in your picker, this retirement barely touches you).

For GPT-5.6 Terra, GitHub offers two successors — GPT-6.1 Sol or GPT-6 Luna — and doesn’t tell you which to pick. That is a real decision, and the right answer depends on your workload. My recommendation: don’t guess from the names. Run your two or three most representative agent tasks once on each replacement, compare the outputs, and standardize on the one that behaves better for your work. The cost of that experiment is an afternoon. The cost of standardizing on the wrong one is months of slightly-worse agent output.

If you want more background on what Opus 5.5 actually brings, our breakdown of the Claude Opus 5.5 launch covers its pricing and positioning, and our guide to GPT-6 Sol and Luna covers OpenAI’s current generation. Both are worth a read before you lock in a choice, because the replacement models have their own pricing and tier rules.

Step 3: Retest your real agent workflows, not just a hello-world prompt

This is the step most migration guides skip, and it’s the one that actually matters. A model swap is not a rename. The replacement model will phrase things differently, make different judgment calls, and handle edge cases differently. A quick “write a Fibonacci function” test proves nothing.

Instead, retest the workflows you actually depend on:

• Your hardest recurring task. Whatever agent workflow you run weekly — the big refactor, the test-suite generation, the migration script — run it end to end on the replacement model and diff the output against your last known-good run. If the output is worse, you need to know now, while you still have time to adjust prompts or instructions.
• Your longest-running agent job. Long agent runs are where model differences compound. A model that drifts slightly on step 3 can be completely off course by step 30. If you have a multi-step agent workflow, watch the full run, not just the first output.
• Your custom agents and skills. If you built custom agents with instructions tuned for Opus 5 or Terra, those instructions were tuned for a specific model’s behavior. The same instructions on a new model can produce different results. Budget time to retune the instructions, not just the model name.
• Your evaluation prompts. If you have any test suite or benchmark prompts you use to judge agent output, run them against the replacement and compare scores. Numbers beat vibes here.

Keep a short note of what changed — “Opus 5.5 writes longer explanations, needs a ‘be concise’ instruction where Opus 5 didn’t” is exactly the kind of finding that saves your team a week of confusion. Share it in whatever channel your team uses for tooling notes.

Step 4: Enterprise and Business admins — check model policies before November 16

If you administer Copilot Business or Copilot Enterprise, your checklist is different from a solo developer’s, and it has a trap in it.

Under GitHub’s default model enablement, the suggested replacements are automatically enabled for Business and Enterprise customers — unless an administrator turned off the global default or explicitly disabled those models. If your organization did either of those things, the replacements will not appear in your developers’ model pickers after November 16, and your developers will be left with a gap instead of a smooth handoff.

Here is the admin checklist:

1. Confirm the replacements are enabled. In GitHub Enterprise Cloud, go to AI controls → Copilot → Configure models. For an organization, go to Settings → Copilot → Models. Verify that Claude Opus 5.5, GPT-6.1 Sol, and GPT-6 Luna are enabled in your model policies. If your org turned off the global default, enable the replacements explicitly.
2. Check multi-org billing. If a developer belongs to multiple organizations, the billing organization — the one listed under “Usage billed to” in Copilot features — dictates which model policies apply. A developer who looks correctly configured in one org can still be governed by another org’s policy. Verify the billing org, not just the visible org.
3. Tell your developers what changed. A one-paragraph internal note — “On Nov 16, Opus 4.8/Opus 5/Terra leave Copilot; Opus 5.5 and GPT-6.1 Sol/Luna are enabled; retest pinned workflows” — prevents a flood of “where did my model go?” messages on a Monday morning.
4. Audit pinned models in shared tooling. Same search as Step 1, but across shared repos, internal tools, and CI. Enterprise teams have more shared surface area, so this matters more, not less.
5. Know your escalation path. GitHub says Enterprise customers with questions or concerns should contact their account team. If your org has compliance requirements around model changes, start that conversation before the deadline, not after.

One reassurance for admins: you do not need to manually purge the outgoing models. GitHub states that no action is required to remove deprecated models after the deadline. Your work is all on the enablement and communication side.

The models aren’t dying — they’re leaving Copilot

This is the reassurance most coverage buries, and it matters if you built something that genuinely needs one of these models. The November 16 date removes the models from GitHub Copilot. It does not delete them from the world.

According to reporting on the announcement, all three models remain available through their vendors’ own APIs after November 16. Claude Opus 5 stays on Anthropic’s API until at least July 24, 2027, and Claude Opus 4.8 until at least May 28, 2027. GPT-5.6 Terra is not on OpenAI’s deprecation list. So if you have a workload that truly depends on Opus 5’s specific behavior — a fine-tuned pipeline, a benchmark you must reproduce, a research setup — you can still call it directly from Anthropic’s API. You just can’t call it through Copilot anymore.

That distinction reframes the whole migration. You are not losing a model; you are losing a convenient way to reach a model. For most developers, switching to the replacement inside Copilot is the right move. For the few who genuinely need the old model, the vendor API is the escape hatch — with its own pricing, keys, and integration work, but it exists.

Why this keeps happening — and how to stop being surprised

This is the third Copilot model retirement wave in about two months. On September 1, 2026, GitHub retired Claude Opus 4.5, Claude Opus 4.6, and Claude Sonnet 4.5. On October 2, it retired Claude Opus 4.7, Gemini 3.5 Flash, Gemini 3.6 Flash, and Kimi K2.7 Code. A September 18 entry set October 19 as the removal date for six more models, including GPT-5.5, GPT-5.4, and Grok 4.5. November 16 is simply the next date on the calendar.

The pattern is structural, not personal. Copilot is a multi-model platform, and model vendors ship new generations every few months. GitHub consolidates its lineup onto the newest generation, retires the old one, and gives developers a few weeks to move. Expect this to keep happening roughly every month or two for the foreseeable future.

So the real lesson of this retirement is not “switch to Opus 5.5.” It’s “build your workflows so the next retirement is boring.” Three habits make that true:

• Never hardcode a model name where a config value will do. Put the model choice in one configuration file, environment variable, or settings entry — not scattered across scripts, prompts, and docs. When the next retirement lands, you change one value and retest, instead of hunting through a dozen files.
• Keep your prompts model-agnostic where you can. Instructions that say “you are an expert code reviewer” survive a model change better than instructions tuned to one model’s quirks. When you do tune for a specific model, label it — a comment like “tuned for Opus 5, retest on switch” saves future-you an hour of archaeology.
• Subscribe to the GitHub Changelog. The deprecation notices land there first, usually a month ahead of the cutoff. A monthly five-minute scan of the changelog is the cheapest insurance against the next surprise.

What to do this week

If you take one thing from this article, make it a date on your calendar. You have until November 16, but the work is small enough to finish this week:

1. Search your repos, configs, and prompt libraries for the three retiring model names.
2. Switch Opus 4.8 and Opus 5 to Opus 5.5; switch Terra to GPT-6.1 Sol or GPT-6 Luna after a quick head-to-head test.
3. Retest your two or three most important agent workflows end to end on the replacement.
4. If you administer Business or Enterprise seats, verify the replacements are enabled in model policies and tell your team.
5. If you genuinely need an old model, plan the move to the vendor’s own API — don’t assume Copilot access will come back.

Model retirements feel like someone moved your tools. They did. But the replacements are already in the picker, the migration is a search-and-retest job rather than a rewrite, and a month is plenty of time — if you start before November 16 starts starting without you.

Frequently Asked Questions

Which GitHub Copilot models are being retired on November 16, 2026?

Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra leave all GitHub Copilot experiences — Chat, inline edits, ask and agent modes, and code completions — on November 16, 2026. GitHub announced the retirement on October 15, 2026.

What replaces Claude Opus 5 in GitHub Copilot?

Claude Opus 5.5 replaces both Claude Opus 4.8 and Claude Opus 5. Note that Opus models require Copilot Pro+ ($39/month) or higher — Copilot Pro ($10/month) does not include Opus models, so Pro users are unaffected by the Opus retirements.

What replaces GPT-5.6 Terra in GitHub Copilot?

GitHub suggests GPT-6.1 Sol or GPT-6 Luna. Try both on your most representative agent tasks before standardizing, since the two models behave differently and GitHub does not designate one as the default successor.

Can I keep using Claude Opus 4.8 after November 16?

Only if you are an individual subscriber on an annual Copilot plan. That exception covers Opus 4.8 alone — not Opus 5 or GPT-5.6 Terra — and does not apply to Copilot Business or Enterprise seats. On request-based billing, Opus 4.8 requests carry a multiplier of 27 premium requests.

What happens to my pinned workflows if I do nothing?

GitHub removes the retired models from the picker automatically, and workflows that name them directly may fail or silently fall back to a default model. Search your configs, custom agents, prompt libraries, and CI jobs for the old model names, switch them, and retest before November 16.

Are Claude Opus 5 and GPT-5.6 Terra being shut down entirely?

No. The November 16 date removes them from GitHub Copilot only. Reporting on the announcement indicates all three remain available through their vendors’ own APIs — Opus 5 on Anthropic’s API until at least July 24, 2027, and Opus 4.8 until at least May 28, 2027 — so workloads that truly need an old model can call it directly.