GitHub Copilot drops Gemini 2.5 Pro and Gemini 3 Flash across every mode

Two Gemini models have disappeared from the Copilot path at once. GitHub says Gemini 2.5 Pro and Gemini 3 Flash are deprecated as of July 31, 2026 across all GitHub Copilot experiences, including Copilot Chat, inline edits, ask and agent modes, and code completions.

The replacement table is short. GitHub points Gemini 2.5 Pro users to Gemini 3.1 Pro (Preview), and Gemini 3 Flash users to Gemini 3.6 Flash. No manual action is required to remove the deprecated models, but teams with workflows, documentation, scripts, or integrations pinned to those model names will need to update them.

The more operational detail sits with Copilot Enterprise. GitHub says administrators may need to enable access to the alternative models through model policies in Copilot settings. Once the policy for a specific model is enabled, the model should appear in the Copilot Chat model selector in VS Code and on github.com. Enterprise customers with concerns are being directed to account managers.

This is a small changelog entry with a broad surface area. Copilot is now an environment that routes developer work across multiple model families rather than a single assistant with one stable backend. When a model leaves that environment, the effects can show up in code completion behavior, agent-mode tool use, inline edit conservatism, latency, and cost controls.

The preview label also matters. Gemini 3.1 Pro (Preview) may be an obvious successor for teams that valued Gemini 2.5 Pro, but some regulated or conservative engineering organizations avoid preview models in standard workflows. Those teams may prefer a stable alternative, or they may need to run their own comparison before changing the default model policy.

The practical checklist is straightforward: search for hard-coded Gemini 2.5 Pro or Gemini 3 Flash references, verify Copilot model policies, and run a focused regression pass on the coding workflows that depend most on model behavior. Model deprecations are less visible than launches, but for organizations standardizing on AI coding tools, they are part of the new maintenance burden.