GitHub deprecated Gemini 2.5 Pro and Gemini 3 Flash across GitHub Copilot on July 31. The removal applies to Copilot Chat, inline edits, ask mode, agent mode, and code completions. GitHub recommends Gemini 3.1 Pro Preview as the alternative to 2.5 Pro and Gemini 3.6 Flash as the replacement for 3 Flash.
What changes for developers
No action is required to remove the old models from the picker. Action is required wherever a saved workflow, instruction, automation, or team policy expects one of the deprecated model IDs. Enterprise administrators may also need to enable the suggested models before users can select them.
| Deprecated model | GitHub’s suggested alternative | Migration caution |
|---|---|---|
| Gemini 2.5 Pro | Gemini 3.1 Pro Preview | Preview availability and behavior may differ |
| Gemini 3 Flash | Gemini 3.6 Flash | Recheck latency, output shape, and agent tool use |
A replacement is not a drop-in guarantee
Model families change instruction following, context behavior, token use, tool selection, and formatting. A workflow that merely compiles after changing the model name can still regress. Re-run a small set of representative tasks before changing the team default.
- One routine edit with tests.
- One repository-scale search and plan.
- One tool-using agent task.
- One long-context review.
- One security-sensitive refusal or escalation case.
Record accepted-task rate, total time, credits or tokens, human repair minutes, and unexpected tool calls. That packet is more useful than arguing from generic benchmark rankings.
What admins should check
Confirm the alternatives are enabled in Copilot model policies, then verify they appear for the affected users. Search shared documentation and automation configuration for pinned model names. If Gemini 3.1 Pro remains a preview in your environment, decide whether a preview model meets the stability requirements of the workflow.
Bottom line
The immediate migration path is clear, but teams should treat it as a controlled model upgrade. The deprecation is another reason to keep prompts, tests, task state, and review processes portable across models.