GitHub has scheduled the October 2, 2026 deprecation of four GitHub Copilot models: Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7. GitHub says the change applies across the Copilot experiences it lists, including Chat, inline edits, ask and agent modes, and completions. To avoid a last-minute switch, migration should be handled on two fronts: choosing the replacement that matches the workload and confirming that the new model is allowed by organization or enterprise policies.
The September 3 announcement already provides a recommended replacement for each model: Gemini 3.8 Flash for both retiring Gemini versions, Kimi K3 for Kimi K2.7 Code, and Claude Opus 5 for Claude Opus 4.7. GitHub also warns that Copilot Business and Copilot Enterprise administrators may need to enable the alternative models in model policies. The primary reference is GitHub's official deprecation changelog.
Which models are retiring and what migration GitHub recommends
| Retiring model | Date | GitHub suggested alternative | Closest usage scenario |
|---|---|---|---|
| Gemini 3.5 Flash | 2026-10-02 | Gemini 3.8 Flash | Fast answers, small edits, syntax questions, and lightweight or repetitive tasks |
| Gemini 3.6 Flash | 2026-10-02 | Gemini 3.8 Flash | Fast answers, small edits, syntax questions, and lightweight or repetitive tasks |
| Kimi K2.7 Code | 2026-10-02 | Kimi K3 | Multi-step agentic workflows and long-context work across larger codebases |
| Claude Opus 4.7 | 2026-10-02 | Claude Opus 5 | Deep reasoning, complex debugging, and multi-step problem analysis |
The usage-scenario column is not based only on model names. Copilot's own comparison documentation classifies Gemini 3.8 Flash as a fast-help option for simple or repetitive tasks, Kimi K3 as an agentic coding and long-context model, and Claude Opus 5 as a deep-reasoning and debugging model. Review the official model comparison before standardizing one choice across a team.
Gemini 3.5 and 3.6 Flash: the most direct move is Gemini 3.8 Flash
For people using Gemini 3.5 Flash or Gemini 3.6 Flash for short tasks, GitHub's suggested replacement is the most straightforward migration in the group. GitHub released Gemini 3.8 Flash in Copilot on September 3 and describes it in the comparison documentation as suited to fast responses and lightweight tasks. That fits workflows focused on low latency: small changes, utility functions, syntax questions, short explanations, and quick prototypes.
In Copilot Business and Enterprise, Gemini 3.8 Flash may already be available when the default enablement policy for released models is active. In the model availability announcement, GitHub says new models are enabled automatically under default model enablement unless an administrator has disabled that global default or explicitly disabled the model. The admin task is therefore not always to “turn on” Gemini 3.8 Flash, but to confirm its actual policy state before the deadline.
Kimi K2.7 Code: Kimi K3 needs extra administrator attention
The move from Kimi K2.7 Code to Kimi K3 deserves additional care. GitHub positions Kimi K3 for multi-step agentic coding and long-context work across large codebases. That can make it a natural evolution for workflows that rely on repository exploration, planning connected changes, and carrying tasks across multiple files.
There is, however, an important governance difference. In Copilot's default availability policy, GitHub lists open-weight models, including Kimi K3, among the models that are disabled by default and are not automatically enabled simply because “Default availability for released models” is on. In an organization or enterprise, administrators should therefore inspect the Models page and explicitly enable Kimi K3 where higher-level policy allows it. The relevant documentation is About default availability of Copilot models.
GitHub also documents model-specific considerations for Kimi K3 and recommends that enterprises evaluate capabilities, limitations, and safeguards against their own requirements. A corporate migration should therefore be more than a find-and-replace of a model name: test real tasks, agentic behavior, security policy fit, and compatibility with the clients the team actually uses.
Claude Opus 4.7: Claude Opus 5 keeps the deep-reasoning focus
For Claude Opus 4.7, GitHub recommends Claude Opus 5. Both sit in the area GitHub associates with deep reasoning and debugging: complex problems, large refactors, architecture decisions, multi-file analysis, and investigation of system behavior.
That does not mean outputs will be identical. A generation change can alter style, latency, credit consumption, tool use, and results for prompts that were already tuned. If a team has prompts, repository instructions, agents, or automations calibrated around Opus 4.7, run a regression set of representative tasks before making Opus 5 the default.
What users should change before October 2
- Inventory every place where the model is pinned. Check Chat model selectors, completion settings, agent instructions, automations, integrations, and any workflow with an explicit reference to one of the four models.
- Switch to the suggested replacement and test in the same client. Availability varies by plan and surface. The supported models in Copilot table is the reference for plan, client, and status.
- Do not assume changing Chat changes inline suggestions. GitHub documents these model selections separately. Changing the Chat model does not change the inline-suggestion model; where the client supports it, the completion model must be configured through its own selector.
- Update clients and extensions. GitHub requires compatible IDE and Copilot extension versions to switch inline-suggestion models.
- Compare quality, latency, and cost. Different models can consume credits differently. A technically compatible substitute may not be the best economic choice for every task.
Checklist for Copilot Business and Enterprise administrators
- Open the organization's or enterprise's Copilot settings and review the Models area.
- Confirm that Gemini 3.8 Flash, Kimi K3, and Claude Opus 5 appear as options compatible with the plan and clients your teams use.
- Check whether a model is enforced as Enabled or Disabled at enterprise level; an organization cannot override a higher-level decision.
- For Gemini 3.8 Flash and Claude Opus 5, verify whether the default availability policy or an explicit setting already grants access.
- For Kimi K3, perform an explicit check: because it is listed among open-weight models, it is not covered by automatic default enablement.
- If users need to choose models manually, make sure organization policy allows model selection on the relevant surfaces.
- Review default-model settings in managed environments. Since September 2, GitHub has allowed any compatible model to be set as a default through enterprise-managed settings; a default that still points to a retiring model should be updated.
- Run a pilot with representative users and document a fallback if a model is unavailable in a particular client.
GitHub's governance instructions are available in Managing the availability of models in an organization and Configuring access to AI models in GitHub Copilot.
How to test the migration without breaking the workflow
Use a small test set that represents the team's real work. For Gemini 3.8 Flash, measure whether small edits, syntax answers, and utility-code generation stay fast and correct. For Kimi K3, test navigation across larger repositories, multi-step agentic tasks, and changes spread over multiple files. For Claude Opus 5, prioritize difficult bugs, architecture decisions, refactors, and analysis of logs or system behavior.
Beyond output quality, record latency, reprompt frequency, credit consumption, tool use, and behavioral differences with existing instructions. The point is to discover before October 2 whether GitHub's suggested replacement is sufficient or whether prompts and procedures need adjustment.
Auto can reduce reliance on a pinned model, but policies still apply
For workflows that do not require a specific provider, the Auto selector can reduce dependence on a model name. GitHub says auto model selection considers the task and availability and can route around degraded or unavailable models. However, Auto only chooses among models supported by the user's plan and permitted by administrative policies. It does not bypass model restrictions or guarantee that a particular family will be used.
Before using Auto as a universal answer to model retirement, check whether the workflow depends on reproducible behavior, predictable cost, a specific model family, or compliance requirements. The official reference is About Copilot auto model selection.
What happens after the deadline
GitHub says no action is required to remove the models after they are deprecated. The required action is the opposite: update workflows and integrations to supported models before October 2, 2026. In organizations with restrictive policies, the critical step is not waiting for users to discover the model is unavailable on the day of the change.
In short: move Gemini 3.5/3.6 Flash to Gemini 3.8 Flash, Kimi K2.7 Code to Kimi K3, and Claude Opus 4.7 to Claude Opus 5; test each replacement in the client actually used; and, for Business or Enterprise, validate model policies with special attention to Kimi K3. That sequence turns a model retirement into a controlled change instead of a workflow interruption.
