Copilot model deprecation: required updates for AI‑assisted development pipelines

Effective 19 October 2026 GitHub removed a set of inference models from all Copilot experiences—Chat, inline edits, ask/agent modes, and code completions. Any workflow that pins a now‑retired model will silently fall back to an unavailable endpoint, so AI engineers, platform teams, and SREs need to verify that their pipelines reference supported models.

What models were retired and what to use instead

• Gemini 3.7 Flash → Gemini 3.8 Flash
• GPT‑5.5 → GPT‑6.1 Sol
• GPT‑5.4 → GPT‑6.1 Sol
• GPT‑5.4 mini → GPT‑6 Luna
• GPT‑5 mini → GPT‑6 Luna
• Grok 4.5 → Grok 4.7

Operational impact on CI/CD and Copilot‑enabled tooling

Most automation that invokes Copilot does so through model identifiers stored in configuration files, environment variables, or secret stores. After the deprecation date those identifiers will no longer resolve, leading to failed completions or stalled chat sessions. Practitioners should:

1. Search codebases for the exact model strings listed above.
2. Replace each occurrence with the suggested alternative.
3. Run a quick smoke test of the affected CI jobs or local tooling to confirm the new model is reachable.

Because the announcement states that no removal action is required, existing references will not cause immediate errors, but they will not benefit from future improvements or bug fixes tied to the new models.

Policy and admin considerations for Enterprise customers

Copilot Enterprise administrators control model access through model‑policy settings. The new models may be disabled by default, so admins must:

• Open the Copilot settings page for their organization.
• Locate the model‑policy section and enable the replacement models.
• Confirm visibility in the model selector inside VS Code or on github.com.

Verification can be done by a simple test prompt in Copilot Chat; the selector should list the newly enabled model. If an organization needs a custom rollout, the policy can be scoped to specific teams or repositories.

Related CloudNinjas coverage: AI engineering.

What This Means For Practitioners

In short, treat the deprecation as a forced upgrade. Update any hard‑coded model names, ensure Enterprise policies grant access to the replacements, and validate the change in a non‑production environment before rolling it out broadly. Keep an eye on future GitHub announcements for additional model lifecycle events, and consider abstracting model selection behind a configuration layer to simplify future migrations.