An old model name can keep accepting requests while the model and billing behind it change.
November 3 update: Google’s November 2 Gemini API release note deprecates Gemini 2.5 Pro and Gemini 2.5 Flash. You can no longer select either in Google AI Studio. Their announced API routing changes take effect on December 1, 2026.
The two legacy models have different destinations
| Legacy API model | Destination from December 1 | Billing after routing |
|---|---|---|
| gemini-2.5-pro | gemini-3.8-flash | Gemini 3.8 Flash rates |
| gemini-2.5-flash | gemini-3.5-flash-lite | Gemini 3.5 Flash-Lite rates |
A migration plan should name the exact source and destination. Saying that all Gemini 2.5 traffic moves to 3.8 Flash would be wrong. Google advises updating the model string and reviewing the relevant migration guide before December 1.
Review the request schema before moving traffic
The release note flags changes to function calling and thought_signature validation. It also replaces thinking_budget with thinking_level and identifies removed sampling parameters and prefilled model turns.
A successful HTTP response does not prove that the application handled every returned tool call correctly. Keep the request, tool response, and final verification result together when evaluating compatibility.
Price a completed task rather than a model name
Google says requests will be billed at the destination model’s rates. Check the current pricing page for the actual model, usage tier, and billing period before calculating a budget. Caching, batch processing, grounding and other tools can have separate effects.
Even with an unchanged unit rate, two task runs may not cost the same. If one run needs twice the output tokens, its output-token component doubles. Retries and human repair can further change the cost of an accepted result.
Use a small compatibility set
Choose representative tasks with results your application can check. Include function calling, structured output, and a failure that requires recovery. Keep prompts, tool schemas, and source data fixed while comparing the old and replacement paths.
Record completion quality, request errors, tool handling, latency, and total cost. Set a rollback threshold before moving traffic. These are proposed migration checks; we have not run a comparative model benchmark.
Google’s October 30 release note also makes Computer Use generally available across the Interactions and GenerateContent APIs. The earlier preview description in this article is therefore superseded. That availability change still needs to be assessed separately from an application’s model migration.
Keep the deadline with the correct guide
Use the Gemini 3.8 guide for the Pro-to-3.8 path and the Flash-Lite migration section for the 2.5 Flash path. The deprecation record provides the lifecycle context.
Our Flash Cyber report covers a separate restricted-access model. Which legacy model strings remain in your scheduled jobs and fallback routes?
Abdessalam Alaoui
I'm Abdessalam Alaoui, founder of Musthave.AI. I write practical guides to AI models, tools, generative media, and ways real people make money with this stuff. No press-release fluff. 20+ years shipping digital products; you'll also find my work around HowTik, Yumasty, and IncomeHelpful.
