Gemini 2.5 Models Retire on Vertex AI on October 20, and Here's What Google Says to Use Instead

Google Cloud's model versions page lists October 20, 2026 as the retirement date for three Gemini 2.5 models on Vertex AI: gemini-2.5-pro, gemini-2.5-flash and gemini-2.5-flash-lite. If your app on Vertex AI still calls any of them, you need to point it at a newer model. Google doesn't say what time of day the retirement takes effect, and the date only applies to Vertex AI. On the Gemini API, which developers use through Google AI Studio, the same three models have no shutdown date, though Google is limiting them to people who already use them.

Google's documentation now calls the service Gemini Enterprise Agent Platform, but the page still sits under the Vertex AI address.

What Google says to use instead

For Gemini 2.5 Pro, Google lists two replacements, gemini-3.8-flash and gemini-3.5-flash. Both are Flash models, and Google doesn't name a Pro model as a replacement. The same table includes gemini-4-argon, released October 13, but it isn't listed as a replacement for 2.5 Pro.

For Gemini 2.5 Flash, the listed options are gemini-3.8-flash, gemini-3.5-flash-lite and gemini-3.1-flash-lite.

For Gemini 2.5 Flash-Lite, Google lists gemini-3.8-flash, gemini-3.1-flash-lite or Gemma 4.

Google says retirement timelines may be extended but won't move earlier than the date listed. When we checked the page on the morning of October 19, Beijing time, the date was still October 20.

The Gemini API is a different story

On Google's Gemini API deprecations page, gemini-2.5-pro, gemini-2.5-flash and gemini-2.5-flash-lite all show no shutdown date announced. Google says the 2.5 models aren't deprecated and will keep being served through the API until further notice. To keep capacity steady, it's limiting access to people who have actively used them before, and it asks new projects to use Gemini 3.5 Flash-Lite or 3.8 Flash instead.

What to do before the cutoff

Google has a full migration guide, but for anyone who just needs to avoid errors when the models retire, its page gives three steps. Point your application at the recommended replacements, test every feature that matters to you, and deploy the change the way you normally would.