GitHub announced on October 15 that it will retire Claude Opus 4.8, Claude Opus 5, and GPT-5.6 Terra across GitHub Copilot on November 16, 2026. The listed replacements are Claude Opus 5.5 for both Claude models, and GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra. For QA teams, this is a practical reason to re-run AI-assisted test, review, and agent workflows before the cutoff—not a reason to assume a newer model produces equivalent output.
What GitHub Copilot is changing
The retirement applies to Copilot Chat, inline edits, ask and agent modes, and code completions. GitHub says the suggested alternatives are automatically enabled for Copilot Business and Enterprise under default model enablement, unless an administrator has disabled the global default or explicitly disabled those models. Claude Opus 4.8 remains available to individual annual-plan subscribers so they retain an Opus option; that exception does not apply to the general retirement announcement.
The operational detail matters: model availability varies by plan and client, and GitHub documents that availability can change. Treat the change as a dependency migration. A prompt that reliably produces a Playwright locator fix, a test-data proposal, or a code-review finding today needs fresh evidence after the target model is selected.
GitHub Copilot model retirement: a QA migration checklist
- Inventory pinned models. Find repositories, IDE settings, shared prompts, Copilot CLI scripts, and team guidance that name any retiring model.
- Check policy availability. An organization may need to allow the suggested replacement before a test can start. Confirm the model picker in each client your team uses.
- Freeze a small regression corpus. Include representative bug reports, flaky-test investigations, pull-request reviews, and safe refactor tasks. Preserve inputs and the human-accepted outcomes.
- Run side-by-side trials. Compare the current and replacement model on the same corpus before November 16. Review correctness, missing edge cases, unsafe suggestions, latency, and credit use.
- Keep deterministic gates independent. Unit, API, UI, security, and lint checks should still decide whether a change can merge. An AI response is evidence to review, not a passing build.
- Record the model in test evidence. Capture model name, client, prompt revision, repository revision, date, and reviewer decision so later regressions are diagnosable.
Why this matters for QA engineers
AI-assisted quality workflows are sensitive to model behavior. A replacement may phrase a defect differently, select different files for a review, or propose a different test boundary. That can be useful, but it means trend lines based on accepted suggestions, defect discovery, or test-generation output are not directly comparable across the switch. Establish a new baseline rather than calling a variance a productivity gain or loss.
A lightweight acceptance test
Use one focused prompt with a known expected review outcome. For example:
Review this pull request for testability risks.
List only findings that cite a file and line range.
For each finding, propose one automated check.
Do not approve the change; flag uncertainty explicitly.Run it against a deliberately seeded pull request and a real, previously reviewed change. A reviewer should score whether the output identifies the seeded issue, avoids invented findings, cites usable evidence, and leaves the final decision to the normal PR checks. Repeat the same test after an IDE or Copilot-extension update, because GitHub’s documentation notes that model access and minimum client versions can differ.
Do not wait for the retirement date
Schedule the evaluation while the retiring model is still selectable. That gives teams a control run, time to resolve policy gaps, and a clean decision record. For the GPT-5.6 Terra replacement, test GPT-6.1 Sol and GPT-6 Luna against the workload you actually operate rather than selecting only on a model name. For the two Claude replacements, validate Claude Opus 5.5 against the same evidence set.
GitHub also cautions that model availability is subject to change and advises users to carefully review and validate code before production use. That is exactly the right posture for a model migration: measured evaluation, documented differences, and the same automated quality gates that protected the release before AI joined the workflow.