GitHub Copilot to Retire Selected AI Models on October 19

On September 18, 2026, GitHub announced that several AI models would be deprecated across Copilot experiences on October 19. The change affects chat, code completion and agent workflows.

Models listed include Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5. GitHub suggests newer alternatives including Gemini 3.8 Flash, GPT-5.6 variants and Grok 4.6.

A MODEL RETIREMENT, NOT THE END OF COPILOT: GitHub Copilot offers several model choices. The announcement concerns selected models scheduled for removal, not the retirement of the Copilot service itself. Users should distinguish a change in model availability from a change in the overall product.

WHICH MODELS ARE AFFECTED: The list includes Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5. Teams should identify whether these names appear in personal settings, internal documentation or automated workflows.

WHAT TO CHECK BEFORE OCTOBER 19: Review not only the model selected in chat but also organization-level permissions, editor extensions, CLI settings and agent configurations. Compare the names visible in Copilot with those embedded in internal documentation to identify migration gaps.

USER AND ADMINISTRATOR RESPONSIBILITIES: Individual developers can inventory the models they select and the tasks they use them for. Administrators should verify which replacements are permitted by subscription and policy. Teams relying on model-specific guidance should coordinate documentation and training updates with IT.

DESIGNING A MIGRATION TEST: Select representative tasks from a real codebase, including short explanations, complex debugging, test generation and multi-file edits. Define success criteria in advance and measure compilation, test outcomes, review corrections and latency. Subjective preference alone is not a sufficient acceptance test.

BEHAVIOR MAY CHANGE: The same prompt can produce different explanation lengths, edit scopes and tool-use patterns after a model switch. In agent workflows, teams should check for unnecessary file changes, safe failure behavior and traceable results. A newer model name does not by itself guarantee better quality or security.

AFTER THE SWITCH: During the first days of migration, watch for model-selection errors, quality complaints and unexpected usage changes. Agree on an alternative model or manual workflow in advance so teams can continue working if the preferred replacement proves unsuitable.

ALTERNATIVES ARE NOT IDENTICAL: GitHub points to newer options including Gemini 3.8 Flash, GPT-5.6 variants and Grok 4.6. A replacement may differ in reasoning, coding style, speed or supported capabilities. Choosing a new model should depend on the actual development task.

CHAT WORKFLOWS: Developers use Copilot Chat for explanations, debugging and design discussions. A model change may affect the structure and reliability of answers. Repeating a representative set of existing questions can reveal differences that matter to a team.

CODE COMPLETION: Completion quality involves more than generating syntactically valid code. Developers also care about response time, irrelevant suggestions and how often proposed code is accepted. Tests should use the programming languages and frameworks that teams work with every day.

AGENT MODE: Agents may plan work, call tools and edit multiple files. Different models can make different choices about these steps. Migration testing should check whether the agent changes unnecessary files, runs appropriate tests and produces a reviewable result.

FIXED MODEL REFERENCES: A model name may be embedded in a workflow, configuration or internal guide rather than selected interactively. Teams should search for references to retiring models and update both executable settings and instructions used by developers.

ORGANIZATIONAL POLICIES: Enterprise administrators may control which models employees can select. An announced alternative may not be available until the organization enables it. Development teams and administrators should coordinate before the scheduled change.

COST AND USAGE LIMITS: Different models may have different usage rules or billing implications. Evaluating only answer quality could overlook operational constraints. Organizations should review their subscription terms and current limits before committing to a replacement.

KEEP A COMPARISON SET: Representative tasks such as explaining code, writing unit tests and refactoring can form a useful migration test. Teams can compare correctness, execution results and review time rather than rely entirely on subjective preferences.

THE SCHEDULED DATE: GitHub identified October 19, 2026 as the planned retirement date. Schedules can change. Administrators should check the latest changelog and actual model availability rather than assume the announcement alone confirms the current state.

A PRACTICAL MIGRATION: Inventory current selections, choose candidates, test representative tasks and review organizational policies. Update documentation and tell users how to switch. Follow-up checks after migration can catch problems that were not visible in initial tests.

Enterprise administrators may need to enable replacement models depending on their policies. Teams using fixed model names in workflows or documentation should review their settings before the scheduled retirement.