Wain AI/Tech Blog

AI news and trends worldwide, updated nearly every day

Gemini API's gemini-2.5-pro and flash Switch to Successors Automatically on December 1 - Billing Follows the Successor, and the Migration Has Breaking Changes

Gemini API's gemini-2.5-pro and flash Switch to Successors Automatically on December 1 - Billing Follows the Successor, and the Migration Has Breaking Changes

In its November 2 changelog entry, Google deprecated gemini-2.5-pro and gemini-2.5-flash. From December 1, requests go automatically to gemini-3.8-flash and gemini-3.5-flash-lite and are billed at those rates. Breaking changes include replacing thinking_budget.

In the November 2, 2026 entry of the Gemini API changelog, Google marked gemini-2.5-pro and gemini-2.5-flash as deprecated1. Starting December 1, any request naming either model is sent to a different model and charged at that model’s price.

Current model nameDestination from December 1
gemini-2.5-progemini-3.8-flash
gemini-2.5-flashgemini-3.5-flash-lite

Requests will reach the destination even if the model name in your code stays the same, but Google states that the move involves breaking changes and tells developers to change the name and read the migration guides before December 11. Neither model can be selected in Google AI Studio any longer. gemini-2.5-flash-lite is outside this change and has no announced shutdown date2.

In September, Google said these were not deprecated

The deprecations table lists June 17, 2025 as the release date for both models2. General availability of 2.5 Pro was covered at the time. In its September 18, 2026 changelog entry, Google restricted the 2.5 family to people who had used it before, said these models were not deprecated and would stay available through the API for now, and pointed new projects to 3.5 Flash-Lite or 3.8 Flash1. The latest notice came about six weeks after that.

Earlier deprecations of gemini-3.7-flash and gemini-3.5-flash also sent calls to successors, though the shutdown column then read as unannounced. This time the table carries a firm date: December 12.

Prices move differently for each pair

Paid-tier Standard prices per 1 million tokens, from the pricing page3:

ModelInputOutput (thinking included)
gemini-2.5-pro (prompts up to 200k tokens)$1.25$10.00
gemini-2.5-pro (prompts over 200k tokens)$2.50$15.00
gemini-3.8-flash (through December 31, 2026)$0.75$3.75
gemini-3.8-flash (from January 1, 2027)$1.50$7.50
gemini-2.5-flash$0.30 (audio $1.00)$2.50
gemini-3.5-flash-lite$0.30 (including audio)$2.50

For the Pro-to-3.8 Flash pair, prompts up to 200k tokens see both rates fall for the single month of December. The 3.8 Flash introductory price ends with the year, so from January 2027 input in this bracket goes from $1.25 to $1.50, while output drops from $10.00 to $7.50. In the over-200k bracket, both rates stay below 2.5 Pro’s even after January. Rates alone do not settle the total bill. Google says 3.8 Flash may by design consume more tokens on long, complicated work, and suggests lowering the reasoning level for everyday jobs4.

For the 2.5 Flash-to-3.5 Flash-Lite pair, rates match, and only audio input gets cheaper.

Grounding with Google Search is counted differently as well3. On the paid tier, 2.5 Pro and 2.5 Flash include 1,500 free requests a day and then charge $35 per 1,000. Both destination models instead include 5,000 free requests a month, shared among Gemini 3 and later models, then charge $14 per 1,000, with a cap of 1,500 a day. The free tier differs too: 2.5 Flash offers up to 500 grounded requests a day there, while the free tier of 3.5 Flash-Lite lists grounding as not available through the API (it can be tried in AI Studio).

Where to look in your code

The changelog’s list of breaking changes covers function calling and thought_signature validation, thinking_level taking over from thinking_budget, and the end of sampling settings and of model-turn prefills1. The main spots to edit, drawn from the migration guides:

  • Thinking settings: swap thinking_budget for thinking_level. 3.8 Flash defaults to medium and returns an error on minimal4. 3.5 Flash-Lite defaults to minimal, and Google recommends medium or high for subagents that call tools or run code5
  • Prefills: a request whose final turn is a non-empty model turn gets an HTTP 400. If you relied on prefilling to skip an introduction or to get JSON, move to a system instruction or structured output5
  • Sampling: temperature, top_p, and top_k are deprecated and ignored, and later model generations are expected to reject them with an error5
  • Function calling: on the generateContent API, every FunctionResponse must carry call_id and name4

The documents we checked do not say what happens to a rerouted request that still sends thinking_budget. Changing the model name and testing ahead of time seems safer than finding out on the day of the switch.

Choosing a different model from the destination

Google picked the destinations, so they may not suit a given workload. The 2.5 Flash destination belongs to the Flash-Lite line, which emphasizes low latency and low cost. Google recommends it as a migration path with scores above 2.5 Flash: 18.0% versus 11.0% on HLE and 74.5% versus 63.7% on CharXIV5. Those are Google’s numbers and say nothing about your own prompts. The 3.8 Flash guide notes that work not needing that much verification can lower the reasoning level or use 3.6 Flash, which stays supported4.

Either way, leaving this to automatic routing means not choosing a model yourself. Searching your code and configuration for these two names and deciding a target per workload lets you budget for both the December 1 switch and the January 1 price change.

Sources

  1. Release notes - Google AI for Developers (November 2 and September 18, 2026 entries)
  2. Gemini deprecations - Google AI for Developers (checked November 4, 2026)
  3. Gemini Developer API pricing - Google AI for Developers (checked November 4, 2026)
  4. What’s new in Gemini 3.8 Flash - Google AI for Developers (migration guide)
  5. What’s new in Gemini 3.6 Flash and 3.5 Flash-Lite - Google AI for Developers (migration guide)

We publish the latest AI news nearly every day.

Subscribe via RSS Get new posts the moment they go live.

Search other keywords →