GitHub Copilot Deprecated Models: All 3 Retirement Batches

GitHub Copilot has now run three separate model retirements in 25 days — and if you only saw one headline, you are probably missing at least one batch that affects your setup. On October 19, 2026, six models vanished from the model picker overnight, including GPT-5.5 and Gemini 3.7 Flash. On November 16, 2026, three more follow: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra — with one odd exception for annual personal plans. And back on October 2, a first batch of four models (two Gemini Flash generations, Kimi K2.7 Code, Claude Opus 4.7) was already retired with almost no notice.

This is the consolidated list we could not find anywhere: every batch, every removed model, every official replacement, what GitHub actually did versus what you have to do yourself, and the two admin-side switches (model policies and global default) that decide whether the replacements even appear in your picker. If you just want the table, it is two scrolls down.

Quick answer

• Batch 2 — removed October 19, 2026 (done): Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Grok 4.5. Replacements: Gemini 3.8 Flash, GPT-6.1 Sol (×2), GPT-6 Luna (×2), Grok 4.7.
• Batch 3 — removes November 16, 2026 (announced October 15): Claude Opus 4.8, Claude Opus 5, GPT-5.6 Terra → Claude Opus 5.5, GPT-6.1 Sol, or GPT-6 Luna. Exception: Opus 4.8 stays for individual subscribers on annual plans.
• Batch 1 — removed October 2, 2026 (done): Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, Claude Opus 4.7.
• No action is required to remove anything — GitHub removes the deprecated models itself. Your only job is to notice that your default changed and switch deliberately instead of letting the picker decide.
• Enterprise/Business admins: replacements auto-enable under default model enablement unless the global default or the specific model was disabled — then you flip them on via model policies.
• Not the same as the Codex GPT-5.5 retirement: OpenAI retired GPT-5.5 from Codex sign-in sessions on October 14; API and API-key sessions were unaffected. Different product, same week — our GPT-5.5 deprecation post covers that side.

The full 2026 retirement table

Three batches, newest impact first:

Scope is consistent across all three: the removals apply across most Copilot experiences — Copilot Chat, inline edits, ask and agent modes, and code completions — not just one surface.

Batch 2, executed October 19: what changed and what to do

The October 19 removal is the batch most people felt. GitHub announced it on September 18, 2026 and executed on schedule, per the changelog article "GitHub Copilot models deprecated on October 19, 2026." Per model:

• Gemini 3.7 Flash → Gemini 3.8 Flash. The 3.8 generation already absorbed 3.5 and 3.6 in Batch 1; this was the last Flash holdover. If your completions felt different on October 20, this is why.
• GPT-5.5 → GPT-6.1 Sol. The highest-impact row. GPT-5.5 was a common manual default for chat and agent modes, and Sol is OpenAI's current default-tier model — the same replacement OpenAI chose on the Codex side.
• GPT-5.4 → GPT-6.1 Sol. Same replacement as 5.5; if you were pinned to 5.4 for cost reasons, note that Sol bills at $2/$10 per million tokens via usage-based billing, which is not the mini-tier price you may have been enjoying.
• GPT-5.4 mini and GPT-5 mini → GPT-6 Luna. Both mini tiers collapse into OpenAI's current efficiency model. This is the row that quietly changed your completions bill if you route high-volume completion traffic through minis.
• Grok 4.5 → Grok 4.7. Straight version bump, the only xAI row in any batch.

Two details from the changelog worth keeping: "No action is required to remove the deprecated models" — GitHub removes them from the picker itself — and "Copilot Enterprise administrators may need to enable access to alternative models through their model policies in Copilot settings." If you are on a Business/Enterprise plan and the replacement models did not appear in your picker on October 20, that admin gate is the reason.

Batch 3, coming November 16: the Opus cull and the annual-plan exception

Announced October 15, 2026, effective November 16, 2026, and it is an Anthropic-flavored batch: Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra all leave, with Claude Opus 5.5 absorbing both Opus slots and GPT-6.1 Sol or GPT-6 Luna taking Terra's.

The wrinkle that makes this batch genuinely confusing: Claude Opus 4.8 will remain available to individual GitHub Copilot subscribers on annual plans. GitHub's stated reason is to keep an Opus option for that group. Practically, that means:

• Individual annual subscribers: nothing changes for you on November 16 — Opus 4.8 stays in your picker. You can still switch to Opus 5.5 at will.
• Individual monthly subscribers, Pro+, Business, Enterprise: Opus 4.8 and Opus 5 both disappear on November 16. If you are on monthly and want to keep an Opus tier, the options are switching to an annual plan before the date or moving to Opus 5.5.
• GPT-5.6 Terra users: pick your replacement deliberately. Sol for near-flagship quality, Luna for the efficiency tier — Terra sat in an odd middle spot and both successors are live today.

Under default model enablement, the suggested replacements auto-enable for Copilot Enterprise and Business customers unless an admin has disabled the global default or the specific model — the same mechanics as Batch 2. And again: nothing needs removing; the models simply stop being offered.

Batch 1, October 2: the dress rehearsal

Worth half a section because it sets the pattern. On October 2, 2026, Copilot retired Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7 — immediately, not on a future date. Replacements: Gemini 3.8 Flash (×2), Kimi K3, and Opus 5.5. Enterprise users again needed model policies to see the replacements.

That batch established the two habits this whole post is really about: old models can vanish from the picker without a migration window you will notice, and the replacements may be policy-gated on business plans. Batches 2 and 3 got announcement lead times (32 days and 32 days respectively); Batch 1 effectively did not.

The admin side: model policies and the global default

Both remaining failure modes for "the replacement model isn't in my picker" live in the same place: Copilot settings → model policies (admin-facing). The mechanics, per the changelog articles:

1. Under default model enablement, new models are automatically enabled unless an administrator turned off the global default or explicitly disables the specific model.
2. If the global default is off, admins enable the alternatives via model policies in Copilot settings; once the policy is enabled, the model appears in the selector in VS Code and on github.com.
3. There is nothing to "un-deprecate" — policy only controls the replacements, never the removed models.

If you are a team lead: check your model policies before November 16, not after. The Opus 5.5 / GPT-6.1 Sol / GPT-6 Luna auto-enable can silently change which model your organization's agents route to, and usage-based billing means the bill moves with it.

GPT-5.5 is a special case: three products, three answers

The same model name is involved in two unrelated retirements this month, which is why searches for "gpt 5.5 deprecation date" return such a mess:

• GitHub Copilot: GPT-5.5 removed October 19 (this post, Batch 2) → GPT-6.1 Sol.
• OpenAI Codex: GPT-5.5 retired from Codex for sign-in users on October 14, 2026 → GPT-6.1 Sol for paid plans, GPT-6 Luna for free tiers. Our Codex-side post has the per-plan split.
• OpenAI API: GPT-5.5 was not retired on October 14 — API and API-key Codex sessions were explicitly unaffected. If you call the API directly, today is not your deadline.

Check which surface you are actually on before changing anything. The deprecation date is per product, not per model.

What to do now: a five-minute checklist

1. Open your model picker (VS Code, github.com, or the Copilot app) and look at what is actually selected. If it shows a removed model, your client silently moved you — pick deliberately.
2. Pin your defaults. In Copilot CLI, /config model persists a choice across sessions; in VS Code and the app, set the default model rather than riding Auto if the model matters to your workflow.
3. Re-test your agent flows. Agent modes hit the replacement model with different strengths — a prompt tuned for GPT-5.5's quirks may need one pass against GPT-6.1 Sol.
4. Admins: review model policies and the global default for the Batch 3 replacements before November 16, and confirm Opus 4.8's annual-plan exception applies (or doesn't) to your seat categories.
5. Watch your usage-based bill for one cycle. The mini→Luna and 5.4/5.5→Sol moves change per-token rates; one billing cycle tells you whether to rebalance.

FAQ

Which GitHub Copilot models were deprecated in 2026?

Three batches: October 2 (Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, Claude Opus 4.7), October 19 (Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Grok 4.5), and November 16 upcoming (Claude Opus 4.8, Claude Opus 5, GPT-5.6 Terra — with Opus 4.8 kept for individual annual-plan subscribers).

Why is GitHub Copilot removing so many models?

The pattern across all three batches is consolidation onto each provider's current generation: Gemini 3.x → 3.8, GPT-5.x → GPT-6.1 Sol/Luna, Opus 4.x/5 → Opus 5.5, Grok 4.5 → 4.7. GitHub's changelog gives no collective rationale beyond pointing at the suggested alternatives, but every removed model has a same-family successor already GA in the picker.

Do I need to do anything when a Copilot model is deprecated?

No removal action — GitHub removes the models itself. What you should do: verify which model your picker landed on after a removal, pin deliberate defaults (/config model in Copilot CLI), and re-run your agent workflows against the replacement. Enterprise admins may need to enable replacements via model policies if the global default was disabled.

Why don't I see Claude Opus 5.5 or GPT-6.1 Sol after the deprecation?

Almost certainly the model-policy gate: on Business and Enterprise plans, suggested replacements auto-enable only if the global default and the specific model were not disabled by an admin. An admin can enable them via model policies in Copilot settings; the models then appear in the selector in VS Code and on github.com.

Is Claude Opus 4.8 going away for everyone?

No. Opus 4.8 is deprecated for most Copilot experiences on November 16, 2026, but GitHub keeps it for individual subscribers on annual plans explicitly so that group retains an Opus option. Monthly individual subscribers and business seats lose it on the scheduled date.

What replaced GPT-5.5 in GitHub Copilot?

GPT-6.1 Sol, per the October 19 changelog. The same replacement appears in OpenAI's own Codex retirement (October 14 for sign-in users), so both products converge on Sol — while the OpenAI API continues to serve GPT-5.5 unaffected.

Will more Copilot models be deprecated?

Almost certainly — three batches in 25 days is a cadence, not a cleanup. The practical defense is the same every time: don't hardcode model choices in team docs or CI, check the picker after any Copilot changelog ping, and keep model policies current on business plans.

Related reading

• GPT-5.5 deprecated from Codex — the OpenAI-side retirement, per-plan replacements and dates
• Gemini 4 Argon — the new flagship that landed in Copilot's picker on October 13
• Claude Haiku 5.5 — the other half of Anthropic's 5.5 generation Copilot now routes to
• Copilot CLI tutorial — where /config model and the model picker live
• Copilot HydraFusion guide — the multi-model orchestrator whose pool these retirements reshape
• Antigravity models overview — where the Gemini 3.8 Flash replacement line also lives