GitHub pulled three AI models from Copilot yesterday. Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra are gone — from chat, inline edits, ask mode, agent mode, and code completions. The replacements went live at the same time. The community is treating this as a loss. It is not: Opus 5.5 scores 14 points higher on Terminal-Bench 4.0 than Opus 5. GPT-6.1 Sol solved noticeably more terminal and C# tasks in GitHub’s own testing. If your model picker updated cleanly, you are running better tools today than yesterday.
What Got Removed
Three models are gone as of November 16, 2026, across every Copilot surface — chat, inline edits, ask, agent, and completions:
- Claude Opus 4.8 → replaced by Claude Opus 5.5
- Claude Opus 5 → replaced by Claude Opus 5.5
- GPT-5.6 Terra → replaced by GPT-6.1 Sol (heavy tasks) or GPT-6 Luna (high-volume, low-cost)
GitHub announced this on October 15 — 31 days’ advance notice. One exception: individual Copilot subscribers on annual plans keep Claude Opus 4.8 past this date.
The Enterprise Admin Gotcha
GitHub’s default model policy auto-enables replacement models for Copilot Business and Enterprise — unless your organization disabled the global default model setting, which many teams do for governance reasons. If your users cannot see Opus 5.5 or GPT-6.1 Sol in the model selector right now, that is almost certainly why.
Fix: Go to your Copilot org settings, find Default availability for released models, and enable the policy per model. The options appear in VS Code’s Copilot Chat model picker once the policy is on.
The Replacements Are Better
Opus 5.5 is not a sidegrade. Per Anthropic’s benchmarks: 66.4% on Terminal-Bench 4.0 versus 52.3% for Opus 5. 57.8% on CursorBench 4.0 versus 46.6%. It writes 27.5% less code for the same output and produces 19% fewer bugs. A 200,000-line codebase audit that took 20 hours on Opus 5 runs in three hours on Opus 5.5, with 2.5× fewer tokens.
GPT-6.1 Sol brings a 1M-token context window and Ultrafast mode — up to 8× faster token generation in Codex. GPT-6 Luna is the right pick for high-volume tasks: $0.10 per million input tokens versus Sol’s $2.00. See OpenAI’s model docs for the full spec comparison.
Copilot and Vendor APIs Are on Different Schedules
All three retired Copilot models still work on the vendor APIs. Anthropic lists claude-opus-5 as active with a tentative retirement of “not sooner than July 24, 2027.” OpenAI does not list gpt-5.6-terra as deprecated. If you need continuity for a production agent pipeline, the vendor APIs give you significantly more runway than Copilot does. GitHub’s model lifecycle moves faster — treat the two as separate concerns.
December 14 Is the Next One to Watch
A second retirement wave is set for December 14, 2026: six models including Fable 5, Haiku 4.5, Gemini Argon, Grok 4.6, and Kimi K3. Kimi K3 has no named replacement yet — if your team depends on it, start planning now.
See the full deprecation changelog on GitHub for the official migration targets and timeline.

