GitHub plans to deprecate six models used across selected GitHub Copilot experiences on October 19, 2026. The change affects Copilot Chat, inline edits, ask mode, agent mode, and code completions, making it important for developers and administrators to check both interactive model selections and automation that pins a specific model.
GitHub has published suggested replacements, but they should not be treated as guaranteed drop-in equivalents. Model availability depends on the Copilot plan, client, administrator policy, and timing. GitHub’s documentation also does not define what will happen to every API, SDK, CLI, or integration that explicitly references a model after its retirement.
What changed
According to GitHub’s changelog announcement, the following models are scheduled for deprecation on October 19, 2026:
- Gemini 3.7 Flash: Suggested replacement, Gemini 3.8 Flash.
- GPT-5.5: Suggested replacement, GPT-5.6 Sol.
- GPT-5.4: Suggested replacement, GPT-5.6 Sol.
- GPT-5.4 mini: Suggested replacement, GPT-5.6 Luna.
- GPT-5 mini: Suggested replacement, GPT-5.6 Luna.
- Grok 4.5: Suggested replacement, Grok 4.6.
The announcement presents October 19 as the deprecation date but does not specify an effective time or time zone. It also says no action is required to remove the deprecated models after that date. That statement does not establish whether an existing script, SDK session, CLI run, API request, or other integration containing an explicit reference will be automatically remapped, rejected, or handled through fallback behavior.
Why pinned workflows need attention
Interactive users may be able to select an available replacement in a supported Copilot experience. Automated workflows are less predictable when they contain a model name directly in configuration or code.
Copilot CLI documentation describes several ways to choose a model: the --model option, the COPILOT_MODEL environment variable, persistent settings in settings.json, and custom-agent configuration. The documented precedence is custom agent, command-line option, environment variable, configuration file, and then the CLI default.
That gives teams several places to inspect before the deadline. It does not, however, document the post-retirement result when one of those locations names a model that is no longer available. The same limitation applies to SDK-based applications. GitHub documents model listing, model selection when creating a session, and switching models during a session, but the supplied SDK documentation does not define retirement handling for these six models.
Teams should therefore avoid assuming that a model name will be silently translated to the suggested replacement. The safest approach is to find explicit references, test the replacement, and update configuration deliberately.
Availability varies by plan and client
Replacement availability is not universal. GitHub’s supported-models documentation explains that access can vary according to the Copilot plan, client or surface, administrator policy, and changes over time. Availability may differ between GitHub.com, an IDE, Copilot CLI, the GitHub Copilot app, and other clients.
For Copilot Business and Enterprise, suggested alternatives are automatically enabled under the applicable default model policy unless an administrator has disabled the global default or explicitly disabled the model. When default model enablement is disabled, administrators can manage access through organization or enterprise model policies. Enabled models appear in the Copilot Chat model selector where the relevant experience supports selection.
Administrators should verify the replacement models in the specific plans, organizations, enterprises, IDEs, extensions, and other clients used by their teams. The model selector’s presence in one Copilot surface does not prove that the same model is available everywhere.
The supplied access documentation states that individual Copilot Free and Student users have access only to auto model selection. It does not establish a deprecation-specific outcome for those users. They should not be assumed to have the same model-selection controls as Business or Enterprise users.
Replacements may change behavior and cost
The suggested mappings are useful starting points, not equivalence guarantees. GitHub’s model comparison documentation describes differences among models in task suitability, latency, quality, and capabilities. A replacement may therefore produce different results on coding tasks, agent workflows, or long-context prompts.
Cost is another migration consideration. GitHub documents different model pricing and AI-credit consumption rates, with the actual impact depending on the model, token usage, and Copilot plan. A replacement that performs acceptably in a small test may have different usage or billing implications when deployed across a larger engineering organization.
Teams should test representative prompts and workflows for output quality, latency, context usage, and cost. Security-sensitive code still requires human review; changing models does not create a deprecation-specific security guarantee or replace existing engineering review practices.
What you should do before October 19
- Inventory model references. Search repositories, scripts, CI or automation configuration, environment definitions, CLI commands, custom agents, SDK session setup, and other integration code for Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, and Grok 4.5.
- Map and validate replacements. Use GitHub’s announced mappings as candidates, then confirm that each suggested model is available in the actual Copilot plan and client.
- Test real workloads. Compare representative coding tasks, agent workflows, latency, output quality, context behavior, and cost before changing production or organization-wide defaults.
- Update configuration. Review Copilot CLI flags, the
COPILOT_MODELenvironment variable, settings files, custom agents, and SDK session configuration. Update IDEs, extensions, plugins, CLI versions, and SDK dependencies where replacement support requires it. - Review administrator policies. Business and Enterprise administrators should inspect default availability and organization or enterprise model policies. Confirm that replacement models are enabled for the users and clients that need them.
- Plan for undocumented behavior. Do not rely on automatic fallback or remapping for explicit references after the retirement date. If an Enterprise deployment has questions about its configuration, GitHub advises contacting the account team.
Important caveats
GitHub has announced the deprecation date, but the supplied documentation does not specify the exact time or time zone. It also does not document a universal migration mechanism for API requests, SDK sessions, Copilot CLI executions, or external integrations that continue to name a retired model.
Auto model selection can reduce dependence on individual model names, but the available evidence does not show that it will repair every explicit reference to a retired model. Model access can also change over time, so a successful test should be repeated in the target client and plan before rollout.
Sources
- GitHub Changelog: Upcoming deprecation of selected GitHub Copilot models
- GitHub Copilot supported models
- GitHub Copilot CLI programmatic reference
- Configure access to AI models
- Copilot Enterprise default model availability
- Copilot SDK compatibility documentation
- GitHub Copilot model comparison
- Copilot models and pricing
