Gemini 2.5 Retirement Hits the API on December 1

Gemini 2.5 Retirement Hits the API on December 1

Google is retiring Gemini 2.5 in its API on December 1, and the change is easy to miss. From that date, requests to gemini-2.5-pro and gemini-2.5-flash will automatically go to newer models: 2.5 Pro moves to Gemini 3.8 Flash, and 2.5 Flash moves to Gemini 3.5 Flash-Lite, at those models' prices. Your app will not break. It will quietly start running on a different model, with different behavior and a different bill. That is the part to check.

Google announced the retirement on November 2. The redirect is automatic, which is convenient and dangerous in equal measure. An app that pinned 2.5 Pro for its reasoning quality will get 3.8 Flash instead, and an app that chose 2.5 Flash for its price will get 3.5 Flash-Lite. Both replacements are newer and probably better, but probably is not a test result. If your product depends on a model's specific behavior, a silent swap is a change you should make deliberately, not discover in production.

What else changed around Gemini

The retirement is one piece of a busy month for Google's AI products. Since October 26, new paid API projects store requests by default, which is a privacy setting worth knowing about if you handle user data. Since October 29, you can create, preview, and edit files directly in a Gemini app conversation: documents, spreadsheets, slide decks, PDFs, Markdown, text files, and code.

The same week, Gemini in Google Slides learned to change several slides or an entire deck at once with one prompt. Fill with Gemini, the feature that fills cells in Google Sheets using AI, has shown since October 28 which web sources Gemini used for each cell. Gemini Notebook, the new name for NotebookLM, has worked as a connected app inside the Gemini app since October 23, and it is off by default in Europe.

On October 28, Google announced Gemini 4 Carbon, its new frontier model, which for now only goes to security specialists. The same day, Google brought ads in AI Mode to ten more countries, including Australia, Canada, India, and Singapore. The pattern is clear: Google is shipping features into the places people already work, while the flagship model stays gated.

What to do before December 1

First, find every place your code calls gemini-2.5-pro or gemini-2.5-flash. Search your codebase, your server configs, and any no-code tools or integrations that let you pick a model by name. Third-party tools are the ones people forget, and they are the ones most likely to keep calling 2.5 after the redirect.

Second, test the replacements. Run your real prompts against Gemini 3.8 Flash and Gemini 3.5 Flash-Lite and compare the outputs with what 2.5 produced. Pay attention to reasoning depth, formatting, and refusal behavior, not just speed. If the new model is worse for your use case, you have a month to adjust prompts or pick a different model, and that is a much better position than discovering the difference from a customer complaint.

Third, check the price. The redirect bills at the new models' prices, so your per-request cost will change. Run your typical monthly volume through the new pricing and make sure the budget still works. A model that is cheaper per token but produces longer answers can cost more, and a model that is more expensive per token but answers in fewer tokens can cost less.

Fourth, update your model pins deliberately. Do not leave the 2.5 names in your code and rely on the redirect. Point at the model you actually tested, so the next retirement does not silently change your product again. While you are in there, review the request storage default for your projects, since new paid API projects now store requests by default.

The ONLYOFFICE 10.0 launch showed how a productivity suite handles a major version change with AI kept optional, and the October Play system update showed how Google ships quiet changes that matter. The Gemini 2.5 retirement is the same kind of change: small on the surface, real underneath. December 1 is three weeks away. That is enough time to test, if you start now.