GitHub Copilot Deprecates Gemini 2.5 Pro and Gemini 3 Flash Across All Experiences

GitHub deprecated Gemini 2.5 Pro and Gemini 3 Flash across all GitHub Copilot experiences effective July 31, 2026. The change, announced via GitHub's changelog, removes both Google models from Copilot Chat, inline edits, ask mode, agent mode, and code completions.

GitHub's announcement did not specify a replacement model, migration timeline for existing user configurations, or the exact reason for deprecation. The changelog post confirms only that the models are no longer accessible across the listed Copilot surfaces as of the effective date.

What we know

• Effective date: July 31, 2026
• Models affected: Gemini 2.5 Pro, Gemini 3 Flash
• Scope: All GitHub Copilot experiences, including Copilot Chat, inline edits, ask mode, agent mode, and code completions
• Source: GitHub's official changelog

GitHub has offered multi-model access in Copilot for several years, letting developers select from providers including OpenAI, Anthropic, and Google DeepMind for chat and agent workflows. Deprecating both a flagship reasoning-tier model (Gemini 2.5 Pro) and a fast, lower-cost tier model (Gemini 3 Flash) simultaneously removes Google DeepMind's presence across both ends of Copilot's model lineup, at least temporarily.

The original changelog entry is brief and does not include benchmark comparisons, pricing changes, or statements from Google DeepMind. GitHub has not published a follow-up post clarifying whether a newer Gemini model will backfill these slots in Copilot's model picker.

What this means

For developers who had pinned workflows, IDE extensions, or CI agents to Gemini 2.5 Pro or Gemini 3 Flash inside Copilot, those configurations will fail or silently fall back to a default model starting July 31, 2026. Teams relying on Gemini's long-context handling or cost profile inside Copilot agent mode should audit their setups immediately and switch to a supported model.

This deprecation is notable less for what it changes technically and more for what it signals about the pace of model churn inside developer tools. Copilot's model marketplace has become a fast-moving layer where underlying providers' models are added and removed on timelines set by GitHub and its partners, not always in sync with the model's own lifecycle at the source lab (Google DeepMind has not issued a parallel deprecation notice for Gemini 2.5 Pro or Gemini 3 Flash on its own platforms as of this writing). Developers building on Copilot should treat model availability as unstable and avoid hardcoding dependencies on any single model identifier without a fallback path.