cd /news/ai-products/github-copilot-drops-gemini-2-5-pro-… · home topics ai-products article
[ARTICLE · art-83980] src=byteiota.com ↗ pub= topic=ai-products verified=true sentiment=· neutral

GitHub Copilot Drops Gemini 2.5 Pro and 3 Flash: Migrate Now

GitHub deprecated Gemini 2.5 Pro and Gemini 3 Flash across all Copilot experiences on July 31, with no grace period, marking the sixth model deprecation since January 2026. GitHub recommends migrating to Gemini 3.1 Pro or Gemini 3.6 Flash, the latter offering configurable reasoning and up to 65% token reduction in agentic workflows. The company also shipped enterprise teams model policy targeting in public preview and a default model enablement policy change on the same day.

read4 min views1 publishedAug 2, 2026
GitHub Copilot Drops Gemini 2.5 Pro and 3 Flash: Migrate Now
Image: Byteiota (auto-discovered)

GitHub deprecated Gemini 2.5 Pro and Gemini 3 Flash across all Copilot experiences on July 31 — no grace period, no fallback. If you were using either model in VS Code, JetBrains, Xcode, the Copilot CLI, or Copilot Chat on the web, they are gone. This is the sixth model deprecation in Copilot since January 2026. The pattern is no longer surprising. The pace is.

What to Migrate To #

GitHub’s own July 31 changelog told developers to switch to Gemini 3.1 Pro or Gemini 3.5 Flash. The better answer for most teams is Gemini 3.6 Flash, which shipped in Copilot on July 21 — ten days before the deprecation — and is strictly better than 3.5 Flash. Here is the replacement map:

Deprecated Best Replacement Why
Gemini 2.5 Pro Gemini 3.1 Pro 1M token context, stronger reasoning
Gemini 3 Flash Gemini 3.6 Flash Configurable reasoning, 17% fewer output tokens
Either Auto mode Future-proof, no manual switching needed

Gemini 3.6 Flash added something 3.5 Flash lacked: configurable reasoning effort, similar to how o1-style chain-of-thought works. For agentic tasks, that matters. Google reports up to 65% token reduction in some agentic workflows versus its predecessor, with pricing dropping to $1.50 per million input tokens and $7.50 per million output.

How to Switch #

In VS Code, open Copilot Chat, click the model picker, and select your replacement. In JetBrains (version 1.5.62+), the picker is in Ask, Edit, and Agent modes. Enterprise admins need to enable the model in Settings → Copilot → Model Policies before users can see it in their IDEs.

For the Copilot CLI, the cleanest path is Auto mode. Run gh copilot suggest and select Auto — it routes based on task complexity, honors your admin’s model policies, and automatically avoids deprecated models. There is a known bug: Gemini 3.1 Pro does not appear in the CLI model list even when enabled at the org level (github/copilot-cli #1703). Until that is fixed, Auto mode is the safer choice for CLI users.

JetBrains users have a separate issue: the deprecation warning was never surfaced in the model picker UI (copilot-intellij-feedback #1886). If you use IntelliJ or another JetBrains IDE, you may not have seen this coming.

Six Deprecations in Seven Months #

This is not random churn. GitHub has deprecated Copilot models roughly every four to six weeks since January: legacy Claude and OpenAI models in January, Gemini 3 Pro in March, Claude Sonnet 4 in May, GPT-4.1 and GPT-5.2 in June, and now Gemini 2.5 Pro and Gemini 3 Flash. The lesson is the same each time: do not pin specific model names in scripts, configs, or CI pipelines. Use Auto mode. GitHub’s Auto mode routes across providers based on task complexity, real-time model health, and admin policy — it will not route you to a deprecated model.

Enterprise: Two More Changes Shipped the Same Day #

GitHub shipped two additional enterprise governance features on July 31.

The first is enterprise teams model policy targeting, now in public preview for Copilot Business and Enterprise customers. Until now, model access was controlled at the organization level. The new feature lets admins grant different model access to specific enterprise teams — for example, giving an AI research team Gemini 3.1 Pro while the rest of the organization stays on cheaper models. GitHub has said this is “the first step in a broader shift toward team-level governance.”

The second is a default model enablement policy change announced July 29 and effective August 26. Going forward, generally available models will be enabled by default on Business and Enterprise plans — meaning new models appear without admin action. If your enterprise needs strict model governance, you now need to opt out rather than opt in. Enterprise admins should audit their model policies before August 26.

What to Do Right Now #

  • Open your model picker, replace Gemini 2.5 Pro with Gemini 3.1 Pro, and replace Gemini 3 Flash with Gemini 3.6 Flash — or enable Auto mode and stop tracking this entirely.
  • If on an enterprise plan, verify your admin has enabled the replacement models. The new default enablement policy does not retroactively enable Gemini 3.1 Pro or 3.6 Flash if your admin previously disabled all Gemini models.
  • Check whether the August 26 default enablement policy requires any action for your org’s governance posture.
  • Consider whether enterprise team-level policy targeting is useful for your org’s model access strategy.

The model rotation is not slowing down. Build your Copilot workflows around that fact.

── more in #ai-products 4 stories · sorted by recency
── more on @github 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/github-copilot-drops…] indexed:0 read:4min 2026-08-02 ·