Gemini API reroutes 2.5 Pro and 2.5 Flash calls from 1 December

Gemini API reroutes 2.5 Pro and 2.5 Flash calls from 1 December

2 November 2026: Google has deprecated gemini-2.5-pro and gemini-2.5-flash in the Gemini API, and from 1 December 2026 every request that still names them will be served by a different model and billed at that model's price. Calls to gemini-2.5-pro go to gemini-3.8-flash; calls to gemini-2.5-flash go to gemini-3.5-flash-lite. Both older models have already been removed from the model picker in Google AI Studio.

Not yet independently verified. Google's changelog does not say whether the routed models answer old requests identically; the breaking changes it lists may cause errors in code that still sends the old parameters. We will update this when it can be confirmed, and remove this note.

Google's release notes of 2 November tell developers to change their model strings before 1 December and to work through the migration guides, because the switch is not only a rename.

The breaking changes

For both routes, Google lists the same breaking API changes:

• function calling and thought_signature validation;
• thinking_level replacing thinking_budget;
• removed sampling parameters;
• removed prefilled model turns.

Code written for the 2.5 models that sets a thinking budget, passes sampling parameters the newer models drop, or prefills the model's turn to steer its output is the code most likely to error or behave differently after 1 December. Agent frameworks that replay function-call history without the signatures the newer models check need attention too.

What the bill does

Google's pricing page, read on 3 November, shows how the routing changes the price of the same call on the paid tier, per million tokens:

Requested model	Served by	Input	Output
gemini-2.5-pro	gemini-3.8-flash	$1.25 (up to 200k-token prompts) becomes $0.75	$10.00 becomes $3.75
gemini-2.5-flash	gemini-3.5-flash-lite	$0.30 stays $0.30 for text, image and video	$2.50 stays $2.50

The 3.8 Flash figures are introductory. Google lists them "through December 31, 2026", rising to $1.50 for input and $7.50 for output from 1 January 2027, so a 2.5 Pro workload routed in December pays the low rate for one month and the doubled rate after that. Even at the January price, 3.8 Flash input stays above 2.5 Pro's short-prompt rate and output below it.

A Pro model replaced by a Flash model

The less obvious change is the class of model. A team that picked 2.5 Pro for its reasoning on hard coding or analysis tasks will, after 1 December, be served by a Flash model that Google describes as its most intelligent Flash, built for long-horizon software engineering and agents, unless it chooses a model itself. Teams with evaluation suites should run them against both target models now rather than discover the difference in production.

Our AI model board lists Gemini 2.5 Pro, Gemini 2.5 Flash and Gemini 3.8 Flash among 143 models.

Sources

• Google AI for Developers: Gemini API release notes, 2 November 2026, 2 November 2026
• Google AI for Developers: Gemini Developer API pricing (read 3 November 2026), 3 November 2026