Gemini 2.5 Pro and Gemini 3 Flash Deprecated in GitHub Copilot: What to Switch To
TL;DR
Gemini 2.5 Pro and Gemini 3 Flash are deprecated in GitHub Copilot. What GitHub suggests instead, the October follow-up deprecations, and how to migrate.
GitHub deprecated two Google models across every Copilot experience on July 31, 2026: Gemini 2.5 Pro and Gemini 3 Flash. The change lands in Copilot Chat, inline edits, ask and agent modes, and code completions, and it is effective immediately. The official changelog entry lists Gemini 3.1 Pro (currently in public preview) as the replacement for Gemini 2.5 Pro, and Gemini 3.6 Flash as the replacement for Gemini 3 Flash.
This is the second model-related removal from GitHub this week: GitHub Models was fully retired on July 30. If your team pins models in Copilot, today is a good day to check which ones you are actually calling.
Last updated: October 3, 2026
Update: October 2, 2026 - four more models deprecated, including the August replacement#
GitHub deprecated four more models across every Copilot experience on October 2, 2026: Copilot Chat, inline edits, ask and agent modes, and code completions. One of them is the model this post recommended in August.
| Model | Deprecated | GitHub's suggested alternative |
|---|---|---|
| Gemini 3.5 Flash | 2026-10-02 | Gemini 3.8 Flash |
| Gemini 3.6 Flash | 2026-10-02 | Gemini 3.8 Flash |
| Kimi K2.7 Code | 2026-10-02 | Kimi K3 |
| Claude Opus 4.7 | 2026-10-02 | Claude Opus 5.5 |
- The August advice is out of date. Gemini 3.6 Flash, the replacement named below, is now itself deprecated, and so is Gemini 3.5 Flash. GitHub points both at Gemini 3.8 Flash; our Gemini 3.8 release guide covers it.
- Kimi and Claude users move up a generation. Kimi K2.7 Code goes to Kimi K3 (see Kimi K3 in GitHub Copilot), and Claude Opus 4.7 goes to Claude Opus 5.5 (see the Opus 5.5 release guide).
- The enterprise rule is unchanged. GitHub says Copilot Enterprise administrators may need to enable the alternative models through model policies, and that no action is needed to remove the deprecated ones. Our model policy targeting post explains the controls.
- Read the lineup below as of August 1. The sections that follow describe the Gemini lineup on that date. Use GitHub's supported models documentation for what is available today.
The pattern worth noting: two Google models GitHub recommended as replacements lasted about two months. If you pin a model in Copilot settings or in an enterprise policy, plan on reviewing that pin every time GitHub's changelog lists a deprecation, not once a quarter.
What actually changed#
The deprecation applies to all Copilot plans and surfaces. The models no longer appear in the model selector, and requests that reference them do not route. There is no grandfathering period and no opt-out: "No action is required to remove the deprecated models," the changelog says, because the removal is server-side.
The one place where action is required is Copilot Enterprise. The changelog notes that enterprise administrators may need to enable the replacement models through their model policies in Copilot settings. Gemini 3.1 Pro is a preview model, and preview models are commonly disabled by default in enterprise policies. If a team wants Gemini 3.1 Pro, an admin has to flip that policy on before the model shows up in the VS Code or github.com model selector.
What survives, per the current Copilot supported models documentation, is a Google lineup of three models: Gemini 3.6 Flash and Gemini 3.5 Flash at GA status, plus Gemini 3.1 Pro in public preview. Gemini 2.5 Pro and Gemini 3 Flash are gone from the list entirely.
Why Gemini 3.6 Flash is a fair upgrade#
Google launched Gemini 3.6 Flash on July 21 as a direct efficiency play: 17% fewer output tokens than Gemini 3.5 Flash per task at $1.50 per million input tokens and $7.50 per million output tokens. On DeepSWE it scores 49% versus 37% for Gemini 3.5 Flash, and on MLE Bench it hits 63.9% versus 49.7%. For Copilot users the token efficiency matters twice: fewer output tokens means faster completions and, on usage-based billing plans, fewer premium requests consumed.
The interesting side note is that Gemini 3.5 Flash-Lite, launched the same day at $0.30 per million input and $2.50 per million output tokens, already outscored Gemini 3 Flash on SWE-Bench Pro (54.2% versus 49.6%) and OSWorld-Verified (74.0% versus 65.1%). In other words, the model being removed was already beaten by the cheap tier of the current line. The deprecation is not a downgrade for most workloads.
What this signals about Copilot's model strategy#
Reading the July changes together, GitHub is consolidating hard. GitHub Models retirement removes the neutral API surface that let anyone call any provider's model with a GitHub token. The new enterprise model policy targeting in public preview gives admins fine-grained control over which models their teams can see. And now the two oldest Gemini entries in the Copilot catalog are gone.
The throughline: Copilot is moving from a marketplace of many models to a curated, policy-managed set. For individual developers the practical cost is low, mostly a model-selector change. For teams that built scripts, prompts, or automation around a specific model string, this is a reminder that model names in Copilot are now moving targets. The GitHub Models retirement post covers the API side; this change is the same churn inside the editor.
How to migrate#
The concrete checklist, based on the changelog and the supported-models documentation:
- Open the model selector in VS Code and in Copilot Chat and confirm which Gemini models are actually listed. If you pinned Gemini 2.5 Pro or Gemini 3 Flash, switch to Gemini 3.1 Pro or Gemini 3.6 Flash respectively.
- If you are on Copilot Enterprise and Gemini 3.1 Pro is not visible, have an admin check the model policy in Copilot settings and enable the preview model. The changelog says verification is via the individual Copilot settings page, and that the model appears in the selector once the policy is enabled.
- Search your repo for the model strings, including any in configuration files or team documentation, and update anything that instructs the model to use the old names.
- Run a quick smoke test on your most important assisted workflow with the new model before relying on it. If you automated prompts that depended on Gemini 2.5 Pro's specific behavior, budget a small eval pass.
If you are coming from a broader model access question, the migration checklist in Migrating Off Retired GPT Models in 2026 covers the general playbook: read the retirement table, map old names to new ones, eval before you switch, and keep a provider fallback. The pattern is identical here, just with a shorter runway.
FAQ#
Which Gemini models did GitHub deprecate in Copilot?#
GitHub deprecated Gemini 2.5 Pro and Gemini 3 Flash across every Copilot experience on July 31, 2026, with Gemini 3.1 Pro and Gemini 3.6 Flash as the suggested replacements. On October 2, 2026 GitHub deprecated Gemini 3.5 Flash and Gemini 3.6 Flash as well, pointing both at Gemini 3.8 Flash. Use GitHub's supported models documentation for what is available today.
Do I need to do anything to remove the deprecated models?#
No. The removal is server-side, so no action is required to remove the deprecated models. The one place action may be required is Copilot Enterprise, where administrators may need to enable the alternative models through model policies in Copilot settings before they appear in the model selector.
Why is Gemini 3.1 Pro missing from my Copilot model selector?#
Gemini 3.1 Pro is a public preview model, and preview models are commonly disabled by default in enterprise policies. On Copilot Enterprise, an admin has to enable the preview model in the model policy in Copilot settings, and it appears in the selector once the policy is enabled.
How should I migrate off a deprecated Copilot model?#
Check which models the selector actually lists and switch any pinned model to GitHub's suggested alternative, have an enterprise admin enable the replacement if it is not visible, search your repo and team docs for the old model strings and update them, and run a smoke test on your most important assisted workflow before relying on the new model.
Continue Reading#
- GitHub Models Is Retired: What to Use for Model Access Now - the API-side retirement that landed one day earlier
- Enterprise Teams Model Policy Targeting in Public Preview - how admins control which models a team can use
- Migrating Off Retired GPT Models in 2026 - the general model-deprecation migration playbook
- Vertex AI Shutdowns This Week: Gemini 2.5 on October 20 - Gemini 2.5 Pro leaving Vertex AI, with Google's replacement ids and the code change
- Gemini 3.5 Pro Developer Guide 2026 - what the current Gemini line offers and how to use it
- Claude Fable 5 vs Gemini 3.1 Pro - a head-to-head on the model GitHub suggests for Gemini 2.5 Pro users
- Kimi K3 Is GA in GitHub Copilot: Pricing, Rollout, and What It Means for Model Choice
- GitHub Copilot - current plans, models, and alternatives on one page
- Gemini - the model family, pricing, and alternatives on one page
Sources#
Get the next deep dive like this in your inbox
One email a week on News and the rest of the AI dev stack. Free.
Read next
Kimi K3 Is GA in GitHub Copilot: Pricing, Rollout, and What It Means for Model Choice
Kimi K3 is GA in GitHub Copilot at $3/$15 per million tokens, off by default for Business and Enterprise. Rollout details and what it means for model choice.
7 min readGemini's Agentic Video Understanding Cuts Video Tokens by 88%: How It Works and What Breaks
Gemini agentic video understanding lets Flash models pick the frames, audio and transcripts to inspect, cutting tokens up to 88%. How it works and its limits.
7 min readGitHub Adds Enterprise Team Model Policy Targeting: Admin Control Over Which AI Models Each Team Gets
GitHub Copilot enterprise model policy targeting lets admins set a company-wide model baseline and grant extra models to teams. How the preview works.
5 min readNew here? Start with
Technical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.
