Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- On December 1, gemini-2.5-pro routes to gemini-3.8-flash, and gemini-2.5-flash routes to gemini-3.5-flash-lite; each uses the target model's rates.
- Server-side routing does not mean an old ID still pins the old model or preserves its behavior. Neither model can be selected in AI Studio, and API access is limited to users who previously used the models.
- For 3.8 Flash, replace thinking_budget with thinking_level, remove incompatible parameters and prefilled model turns, and check function calling and thought signatures.
- The thinking_level rules differ for 3.5 Flash-Lite: minimal is the default for high-volume work, while medium or high can suit tool-using agents. gemini-2.5-flash-lite is not deprecated by this notice.
What are the two API routes changing on December 1?
Google's November 2, 2026 Gemini API release notes say that, starting December 1, 2026, API requests using gemini-2.5-pro will route automatically to gemini-3.8-flash and be billed at Gemini 3.8 Flash rates. Requests using gemini-2.5-flash will route to gemini-3.5-flash-lite and be billed at Gemini 3.5 Flash-Lite rates.
This changes server-side routing and billing; the old strings will no longer identify the old models. Google has not given an exact switch time or time zone for December 1. Update explicitly to the target model IDs before that date so the old name does not hide which model is actually running.
Does a successful request with an old ID mean nothing changed?
No. The release note describes automatic routing and target-model rates for requests using the old IDs. Even if an old-ID request returns a response, that does not prove Gemini 2.5 Pro or Flash still generated it, or that model behavior and billing stayed the same.
Google AI Studio no longer offers either 2.5 model for selection, and the deprecation page limits API access to users who actively used the relevant models before. Check availability for your own project, region and API features in the official console and request records; the routing notice is not a promise that every project can use the target models.
What should Pro callers check when moving to Gemini 3.8 Flash?
Changing the model ID to gemini-3.8-flash is only the first step. The 3.8 Flash migration guide replaces thinking_budget with the string enum thinking_level: low, medium (the default) or high. minimal is unsupported. Remove temperature, top_p, top_k, candidate_count, which is unsupported in Gemini 3 and later, and prefilled model turns.
For server-side conversation interfaces that support previous_interaction_id, follow that interface’s migration rules for state and remove prefilled model turns. Do not apply previous_interaction_id to every GenerateContent workflow. If your application reuses legacy request builders, inspect the parameters and turn structure it actually sends before switching traffic. For more detail on 3.8 prices and staged rollout, see the 3.8 Flash pricing and migration guide. This article focuses on the deadline affecting the two old 2.5 IDs.
Check function calling and thought signatures too
The Gemini 3.8 Flash guide says to put multimodal assets inside the response payload and separate inline instructions with two newline characters. If pre-tool text triggers a Malformed_Function_Call error, check the corresponding case in Google's function-calling documentation. With the GenerateContent API, every FunctionResponse must include call_id and name.
Tool calls and multi-turn state on the new models also involve thought_signature validation. Update the SDK and follow Google's signature guidance so the signatures and call-correlation fields needed for tool round trips are preserved. Do not drop them while rebuilding conversation history. Validate field behavior against the API and SDK you use.
Flash callers have different thinking-level guidance
The target model is gemini-3.5-flash-lite. Its guide gives thinking_level minimal as the default: keep it for high-volume extraction, routing or classification. For autonomous subagents that need tool calls, code execution or multi-step reasoning, choose medium or high to reduce premature tool termination.
This differs from 3.8 Flash, which does not support minimal. The 3.5 Flash-Lite guide applies the same Gemini 3.6 Flash API and function-calling rules: remove temperature, top_p, top_k, candidate_count and prefilled model turns, and refer to the thought-signature guidance. Configure and test each target separately instead of copying the Pro path's thinking setting to Flash.
Complete a controlled migration before December 1
Find both old IDs in application code, configuration, scheduled jobs, Batch work, SDKs and third-party gateways; record each caller and owner. Change each caller explicitly to its target ID, then check request validation, prompt formatting, multi-turn state, function calls, response parsing and failure handling in a non-production environment. Compare quality, latency and usage on samples your team has approved.
Budget using each target model's current official rates, then compare them with project usage and bills. This guide does not list unverified amounts or report account-level API tests. Google's deprecation page says gemini-2.5-flash-lite is not deprecated and remains available through the API until further notice; it is a different ID from gemini-3.5-flash-lite, the target for gemini-2.5-flash. The sources do not confirm an exact switch time, region or project-level availability.
Sources
Frequently Asked Questions
Does this notice change Gemini Apps plans too?
That cannot be inferred. The notice concerns two Gemini API model IDs and server-side routing; it does not announce changes to personal Gemini Apps plan benefits, model selection or subscription prices.
What if a third-party gateway still shows an old model name?
Ask the gateway provider which upstream model ID it will target after December 1, which model rate it will bill, and which log fields show that mapping. A UI label alone does not prove which model actually ran. Check request records and bills.
Can changing IP preserve the old model or grant access to the new one?
No. Server-side routing, account eligibility and model rates do not depend on the network exit. Changing IP does not undo automatic routing or grant API permission. Troubleshoot the network only when there is evidence of DNS, TLS or connection failure.