The old model name will reach a different model
Google’s November 2 notice says December 1 API requests for Gemini 2.5 Pro will route to Gemini 3.8 Flash; 2.5 Flash requests will reach Gemini 3.5 Flash-Lite. An API connects an app to a software service. The destination model’s billing applies. Both older models are already removed from AI Studio’s selector. [1]
Allow for settings and price changes
If your app asks AI to use a calendar or another tool, its developer needs to revisit those calls and the settings that control how much reasoning work the model does. The guides also flag thought signatures, validation data that must survive between requests. Flash-Lite ignores some old randomness settings; a request ending with a nonempty prewritten model reply already causes an error. [3][4]
Standard paid prices are in US dollars per million tokens, small pieces of text or other input used to meter usage. Gemini 3.8 Flash input costs $0.75 and output $3.75 through December 31; those rates become $1.50 and $7.50 from January 1, 2027. Gemini 3.5 Flash-Lite lists $0.30 input and $2.50 output. Output charges include the model’s reasoning work; tools and saved-context storage can add charges. These figures do not predict your total bill. [2]
Our take
Treat this as an app change with a deadline. For a made-up booking, give the assistant two available times and an unavailable third option. Check that it chooses a real slot, handles the scheduling tool correctly, and returns every field your app expects.
Record the requested model, completed result, and usage before and after migration. A familiar model name in a settings file does not tell you whether the resulting work still meets your needs. We have not tested these replacements or checked an account’s bill.