GitHub Copilot Retires Opus 5 and GPT-5.6 Terra on November 16

GitHub Copilot is about to lose three of the models in its picker. According to GitHub's October 15, 2026 changelog, Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra will be deprecated across all GitHub Copilot experiences on November 16, 2026 — and that includes agent mode, the increasingly central way developers hand multi-step work to AI.

The deprecation is broader than the chat window. The GitHub changelog entry states that the removal covers Copilot Chat, inline edits, ask and agent modes, and code completions. Teams whose scripts, CI pipelines or custom agents pin one of the three model names by ID should expect those calls to fail or be rerouted once GitHub Copilot pulls them from the selector.

What retires, and what GitHub says to use instead

GitHub Copilot's recommended migration path pairs each retiring model with a newer alternative. Claude Opus 4.8 and Claude Opus 5 are both directed toward Claude Opus 5.5, while GPT-5.6 Terra users are pointed at GPT-6.1 Sol or GPT-6 Luna.

One carve-out survived the cleanup: Claude Opus 4.8 will remain available to individual GitHub Copilot subscribers on annual plans, so those developers keep an Opus option after the cutoff. Everyone else — including monthly individual subscribers and all business tiers — loses all three models on November 16. As Windows Forum reported, the suggested replacements are automatically enabled for Copilot Enterprise and Copilot Business customers under default model enablement, unless an administrator has turned off the global default or explicitly disabled the model. Organizations that customized their model policies can enable access through Copilot settings before the deadline.

GitHub's message to administrators is direct: update workflows and integrations to the supported models before the date. The company adds that no action is required to remove the models afterward — they simply disappear from the selector.

The catch for agent workflows

The deprecation lands hardest on teams that automated around specific model names. A migration guide published on Medium advised developers to migrate explicit model selections before November 16, search scripts, CI configurations and custom agents for hardcoded model references, and retest real agent workflows against the replacements before cutover rather than after.

There is an important nuance: none of the three models is being retired by the company that built it. According to a breakdown on DEV Community, Anthropic's own deprecations page lists claude-opus-5 as active with retirement not sooner than July 24, 2027, and claude-opus-4-8 active until not sooner than May 28, 2027, with the company's policy of at least 60 days' notice before retiring a released model. OpenAI's deprecations page does not list gpt-5.6-terra at all — it appears there only as the recommended replacement for older models such as gpt-5-mini, which leaves on December 11, 2026.

After November 16, in other words, all three models remain callable through their makers' own APIs. What closes is one particular route to them: the GitHub Copilot picker. For developers who chose Copilot precisely so they would not have to manage model access themselves, that distinction matters less than the migration work in front of them.

A rotation with no end in sight

The mid-November cleanup continues a rotation that has been running all year. Windows Forum's reporting notes that GitHub previously retired Claude Opus 4.7, Gemini 3.5 Flash, Gemini 3.6 Flash and Kimi K2.7 Code on October 2, 2026, following an earlier September 1 wave that phased out Claude Opus 4.5, Claude Opus 4.6 and Claude Sonnet 4.5. The pace is roughly one retirement wave every few weeks.

Copilot Studio is moving on a parallel track. A Microsoft 365 message-center notice tracked by the Tophhie Cloud blog lists its own set of model migrations on November 12, moving agents off GPT-4.1, GPT-4.1 Mini, GPT-5 Chat, GPT-5 Auto and GPT-5 Reasoning onto replacement models — automatically, whether agent owners are ready or not.

The lesson for teams building on coding agents is structural. Model names in the Copilot ecosystem have become short-lived configuration, not stable dependencies. Pipelines that hardcode a specific model ID inherit every deprecation wave as tech debt, while pipelines that test replacement models against real development work — rather than generic prompts — absorb each rotation with a retest instead of a rewrite. With GitHub Copilot now cycling models roughly monthly, that discipline is becoming a basic cost of doing agent-assisted development.