GitHub announced on July 31, 2026 that Gemini 2.5 Pro and Gemini 3 Flash had been deprecated across all GitHub Copilot experiences. The scope includes Copilot Chat, inline edits, ask mode, agent mode, and code completions. This is therefore more than a chat model-selector change: it reaches the coding and agent workflows inside an IDE.
GitHub recommends Gemini 3.1 Pro (Preview) as the replacement for Gemini 2.5 Pro and Gemini 3.6 Flash as the replacement for Gemini 3 Flash. The company says Copilot Enterprise administrators may need to enable the alternatives through model policies in Copilot settings and confirm that the models are visible to users. No separate administrator action is required to remove the deprecated models.
The real migration cost is usually not selecting a replacement. It is the hidden dependency that an existing workflow has on a model identifier, output format, or context behavior. Teams using agent mode may have fixed tool permissions, prompts, regression prompts, or approval steps. After a model change, the feature can remain available while output length, tool selection, coding style, and failure modes change.
Enterprises should treat model identifiers as governed configuration rather than values embedded in runbooks or automation scripts. Inventory the affected Copilot policies, IDE settings, onboarding material, and internal evaluations, then create a small regression set for each replacement. Gemini 3.1 Pro is still a Preview, so teams should not assume equivalence with the retired Gemini 2.5 Pro. High-impact work still needs human review, version records, and a fallback path.
The announcement also shows that model supply has become a platform-management problem. When Copilot spans chat, completions, and agent mode, AI governance has to cover more than who may use Copilot. It also has to cover which models a team permits, when a model may be replaced, and how the replacement is validated. Model policies, observability, and regression testing directly affect whether a development workflow can transition cleanly.
For developers, the immediate check is whether any existing workflow directly specifies the two retired model identifiers, whether the replacement models are enabled by the organization policy, and whether high-risk agent tasks still behave as expected. GitHub does not require extra cleanup for the old models, but enterprises still need to update integrations and tests so platform model lifecycle changes do not become an unexpected production interruption.
