GitHub's changelog confirmed that, as of July 31, 2026, Gemini 2.5 Pro and Gemini 3 Flash are deprecated across every GitHub Copilot surface — Copilot Chat, inline edits, ask/agent modes, and code completions. This is a routine-looking changelog entry that has a real practical bite for any team that pinned a Copilot workflow, custom instruction set, or CI integration to one of these specific models: those integrations will stop working as expected unless an admin updates model policies to point at a currently supported model. It's also a useful reminder of a broader operational reality that's easy to ignore when picking a model for a workflow: unlike open-weights models a team hosts itself, models accessed through a managed coding-agent platform are subject to the platform's own lifecycle, and that lifecycle moves fast right now, since frontier labs are shipping new model generations every few weeks. For engineering teams, the actionable implication is to treat "which model powers our Copilot workflows" as a piece of infrastructure that needs monitoring and a migration plan, the same way you'd track a deprecated API version or an EOL runtime, rather than a "set once" configuration. It's also a useful signal about where the model-quality frontier is moving inside Copilot specifically: as older Gemini versions get retired, the practical default choices left for most teams narrow toward whatever the current-generation options are (newer Gemini, GPT, and Anthropic models supported in the platform), which matters if you've built prompts, instructions, or evaluation harnesses that were tuned against the behavior of a specific now-deprecated model and may need re-tuning against its replacement.