GitHub announced on September 3, 2026 that four more AI models will leave GitHub Copilot on October 2, 2026. The change applies across Copilot Chat, inline edits, ask and agent modes, and code completions.
This is a separate retirement wave from the models removed on September 1. Teams using named models in test-generation prompts, coding-agent procedures, training material, or comparison dashboards should treat October 2 as a migration deadline.
Models retiring and GitHub’s suggested alternatives
| Retiring model | Suggested alternative |
|---|---|
| Gemini 3.5 Flash | Gemini 3.8 Flash |
| Gemini 3.6 Flash | Gemini 3.8 Flash |
| Kimi K2.7 Code | Kimi K3 |
| Claude Opus 4.7 | Claude Opus 5 |
GitHub says no action is needed to remove the old models after retirement. However, Copilot Business and Enterprise administrators may need to enable the replacements in organization or enterprise model policies before users can select them.
Why this matters for QA engineers
A replacement model can produce different tests, assertions, selectors, patches, tool-call sequences, and failure explanations from the same prompt. A model-picker update is therefore a test dependency change, especially when Copilot agents can edit files and run commands.
GitHub’s supported-model documentation also notes that availability depends on the Copilot plan and client. Some recent models require newer IDE or extension versions, while organization policies can restrict what appears to users. A migration that works in one editor may still fail in another surface or managed account.
A focused four-week QA migration plan
- Find dependencies: search team instructions, saved prompts, screenshots, onboarding guides, evaluation jobs, and automation wrappers for the four retiring model names.
- Confirm policy access: verify that each suggested replacement is enabled for the intended organization, plan, and Copilot client. Test both an allowed user and a deliberately restricted user.
- Run paired evaluations: compare the retiring and replacement model on a fixed set of real QA tasks before October 2. Include test generation, a flaky-test diagnosis, an API-contract change, and a multi-file repair.
- Score outcomes, not prose: measure whether tests run, defects are reproduced, assertions detect the intended failure, unrelated files stay unchanged, and the final patch survives human review.
- Exercise agent guardrails: repeat one task with a failed command, missing dependency, denied permission, and misleading test output. Verify that the replacement stops safely and reports evidence instead of claiming success.
- Record the new baseline: capture model, client version, repository commit, prompt, permissions, tool calls, latency, credit usage, and test results. Update runbooks only after the replacement meets the agreed threshold.
Do not assume the suggested model is behaviorally identical
GitHub’s alternatives are migration recommendations, not compatibility guarantees. Gemini 3.8 Flash, Kimi K3, and Claude Opus 5 may differ in context handling, reasoning, cost, latency, and supported capabilities. Keep prompts and repository state fixed during comparison, then choose the replacement that passes your team’s evidence-based acceptance criteria.
Bottom line
The October 2 retirement gives QA teams a short but usable window to migrate deliberately. Inventory named-model dependencies now, validate policy and client availability, and run a small regression suite before changing the team’s default.
Sources
- GitHub Changelog: Upcoming deprecation of selected GitHub Copilot models (September 3, 2026).
- GitHub Docs: Supported AI models in GitHub Copilot (accessed September 4, 2026).