Gemini 2.5 Pro と Gemini 3 Flash が非推奨にGemini 2.5 Pro and Gemini 3 Flash deprecated

GitHub has begun retiring two of Google's Gemini models from its Copilot platform, a move that will require developers who rely on them to switch to other options. As of July 31, 2026, the company deprecated Gemini 2.5 Pro and Gemini 3 Flash across all GitHub Copilot experiences, according to a post on the GitHub changelog. The change matters because it affects every surface where these models were available, and because it reflects the ongoing churn of a product that now routes requests through a rotating roster of large language models.

The deprecation is broad in scope. GitHub says the affected models are being removed from Copilot Chat, inline edits, the ask and agent modes, and code completions. In practice, that means users who had explicitly selected either model, or who had come to depend on their behavior in day-to-day coding, will need to migrate to an alternative before the models are fully removed. Deprecation typically signals the first stage of a retirement process: the models are no longer recommended and may stop being selectable, ahead of an eventual full shutdown. GitHub's changelog entries of this kind usually point affected users toward comparable successors and outline how model pickers will behave once the transition takes effect.

For context, GitHub Copilot moved away from a single-model design over the past few years and now offers a menu of models from multiple providers, including Anthropic's Claude family, OpenAI's GPT series, and Google's Gemini line. That multi-model approach gives users flexibility to match a model to a task, but it also means individual models cycle in and out as vendors release newer versions and older ones reach end of life. Gemini 2.5 Pro was positioned as a higher-capability reasoning model, while the Flash branding across Google's Gemini lineup has generally denoted lighter, faster, and lower-cost variants aimed at latency-sensitive or high-volume work. Retiring both a Pro and a Flash tier at once suggests GitHub is consolidating around newer Gemini releases or other models rather than dropping Google support altogether, though the changelog framing centers on the deprecation itself rather than a specific replacement.

Model deprecation has become a routine part of the generative AI tooling landscape, and it carries practical consequences that are easy to overlook. Teams that have tuned prompts, built internal workflows, or standardized on a particular model's output style can see subtle changes when they are forced to switch, since different models vary in tone, verbosity, instruction-following, and coding accuracy. Organizations with compliance or reproducibility requirements may also need to document the change. For that reason, the timing of a deprecation notice—and the window before full removal—tends to be the most important detail for enterprise users, even when the headline is simply that a model is going away.

The move also fits a wider industry pattern in which AI providers retire older checkpoints relatively quickly to steer usage toward current models. Google, OpenAI, Anthropic, and others regularly publish deprecation schedules for their APIs, and downstream platforms like GitHub Copilot inherit those timelines while managing their own product decisions on top. Because Copilot abstracts the underlying model behind a unified interface, GitHub can add or remove options without changing the core developer experience, but the tradeoff is that users have limited control over how long any given model remains available.

Developers who used Gemini 2.5 Pro or Gemini 3 Flash in Copilot should check the model selector in their editor or the Copilot settings to confirm which alternatives are now offered, and review any automation or scripts that reference the deprecated models by name. Those on GitHub Copilot Business or Enterprise plans may also want to verify whether administrators have restricted the available model set, since organization-level policies can determine which successors are enabled. As with previous Copilot model changes, the most reliable source for exact migration guidance and any remaining availability window is the official GitHub changelog entry that announced the deprecation, which appeared on The GitHub Blog.