Gemini 2.5 retires on 1 December: what should you do?

In short

• From 1 December 2026, Google will automatically route requests to gemini-2.5-pro and gemini-2.5-flash in the Gemini API to Gemini 3.8 Flash and Gemini 3.5 Flash-Lite. Your integration keeps answering, but with a different model and possibly with errors.
• The switch brings changes that can break existing code, including in function calling, and the price changes with it. Since 26 October, new paid projects also store requests for 55 days by default.
• This week, find out which integrations use Gemini 2.5, test the successor on your own examples and switch yourself before 1 December, so that you decide when it changes.

On 2 November 2026 Google announced that gemini-2.5-pro and gemini-2.5-flash are deprecated in the Gemini API. From 1 December, all requests to 2.5 Pro will automatically go to Gemini 3.8 Flash, and requests to 2.5 Flash to Gemini 3.5 Flash-Lite, at those models' prices (Google AI Studio). The Gemini API is the connection software uses to talk to Google's models, for example a chatbot on your website or an automation in n8n, Make or Zapier.

Unlike OpenAI retiring GPT-4o, your integration will not simply stop on 1 December. That is exactly what makes it risky. A different model quietly takes over, and Google itself lists "breaking API changes" that can make existing code fail. If you switch and test now, you choose the moment. If you wait, you find out when a customer gets an odd answer.

Who is affected?

Anything that calls one of these two models with a Google AI Studio API key:

• a chatbot or form on your website that uses Gemini
• automations with a Gemini step
• your own software, a client portal or an internal tool with a Gemini integration
• a supplier's software that uses Gemini 2.5 under the hood

You are not affected if you only use the Gemini app or Gemini in Workspace, such as in Gmail or Docs. If your integration runs through Vertex AI, Google Cloud's AI platform, a different schedule applies: there the retirement date for Gemini 2.5 Pro, 2.5 Flash and 2.5 Flash-Lite was already 20 October 2026 (Google Cloud). If you use Vertex AI and have not noticed anything yet, check today.

What changes with the automatic switch?

The model. Gemini 2.5 Pro is a Pro model; its successor is a Flash model. Google does not name a new Pro model as the replacement for 2.5 Pro. A Flash model is faster and cheaper, but whether it does your work as well, you only know after testing it on your own examples.

The code. Google lists three kinds of breaking change (Google AI Studio):

• function calling, which lets the model trigger an action in your software, and thought_signature validation
• thinking_level replaces thinking_budget as the setting for how long the model thinks
• some randomness settings are removed, as are prefilled model turns in the prompt

If your integration uses any of these, it may start returning errors after 1 December, even though the request still names the old model.

The price. You pay the new model's price (Google AI Studio):

The 2.5 Pro prices apply to requests up to 200,000 tokens. Coming from 2.5 Pro, you pay less in December, but from January you pay more for input than you do now. Work that through if you process a lot of long documents.

Why should you check storage as well?

Since 26 October, new paid projects using the Generate Content API store their requests and responses by default, with store=true. Projects created before that date keep their current setting (Google AI Studio). Stored requests appear in the AI Studio logs and are deleted after 55 days by default; you can shorten that to 7, 14 or 28 days. Logs you save to a dataset do not expire (Google AI Studio).

That matters if you create a new project for the switch. If personal data passes through your integration, such as names or customer questions, it will sit for 55 days in a log you never deliberately switched on. Turn storage off in the log settings, or per request with store=false, if you do not need it. According to Google, the paid tier does not use your content to improve its products; the free tier does (Google AI Studio). What this means for your retention policy is covered in Retaining AI conversations.

How do you switch before 1 December?

1. List your integrations. Check in AI Studio which projects and API keys exist, and note which tool uses each key. Have a developer search code and configuration for "gemini-2.5". Ask suppliers which model they use and when they will switch.
2. Name an owner for each integration. One person tests, approves the switch and checks the output afterwards.
3. Collect twenty to fifty real examples. Include ordinary and difficult cases from recent months, with output that was right.
4. Test the successor. Run the examples on gemini-3.8-flash or gemini-3.5-flash-lite and compare content, tone, length and format. If a next step reads the output automatically, test that step too. If 2.5 Pro work does not hold up on Flash, compare it with a heavier model from Google or another provider.
5. Update the code. Replace thinking_budget with thinking_level, retest function calling and remove settings that no longer exist. Update the SDK if you want to set store per request; for Python that means google-genai 2.33.0 or later (Google AI Studio).
6. Check storage and costs. See whether the project stores requests, and use a week of real usage to work out the January price.
7. Switch yourself and monitor for a week. Put the new model name in your integration well before 1 December and spot-check output and errors during the first week.

The same approach works for the OpenAI models retiring on 14 January 2027. If you also run GPT-4o or GPT-4.1, handle both in one round: Will your AI integrations stop on 14 January?

When can you ignore this?

• You only use the Gemini app or Gemini in Workspace, with no integration of your own.
• Your integrations already run on Gemini 3.5 or later. In that case, only check the storage setting of new projects.
• Your supplier has confirmed in writing that they will switch and test before 1 December.
• It is an old experiment nobody uses. Then revoke the API key.

Google's other news this month is in Google AI news November 2026.

Sources

• Google AI Studio, Changelog, 26 October and 2 November 2026, accessed 5 November 2026
• Google AI Studio, Pricing, accessed 5 November 2026
• Google AI Studio, Logs and datasets, accessed 5 November 2026
• Google Cloud, Model versions and lifecycle, accessed 5 November 2026