Gemini API will reroute Gemini 2.5 Pro and 2.5 Flash calls on December 1

13 hours ago

7 min read

Google has deprecated Gemini 2.5 Pro and Gemini 2.5 Flash on the Gemini Developer API, and from December 1, 2026 every request that still names either model will be served by a newer model and billed at that model's rates. Calls to `gemini-2.5-pro` will route automatically to `gemini-3.8-flash`, and calls to `gemini-2.5-flash` will route to `gemini-3.5-flash-lite`. The change was published in the Gemini API release notes on November 2, 2026, and the deprecations page now lists December 1, 2026 as the shutdown date for both models.

For most Gemini 2.5 Pro workloads the bill goes down in December and then rises again in January. Gemini 3.8 Flash is listed at $0.75 per 1M input tokens and $3.75 per 1M output tokens through December 31, 2026, against $1.25 and $10.00 for Gemini 2.5 Pro on prompts up to 200,000 tokens. From January 1, 2027 the official Gemini 3.8 Flash rates double to $1.50 and $7.50, which puts input above the old Gemini 2.5 Pro price while output stays cheaper. For Gemini 2.5 Flash users the standard text price does not move at all, since Gemini 3.5 Flash-Lite charges the same $0.30 input and $2.50 output, and it is cheaper for audio input.

Both 2.5 models are already gone from the model picker in Google AI Studio. API access is limited to existing users who have actively used them in the past, so new projects cannot start on them. Gemini 2.5 Flash-Lite is unaffected and, according to the deprecations page, remains available through the API until further notice.

··········

GEMINI 2.5 PRO AND 2.5 FLASH DEPRECATION: WHAT GOOGLE PUBLISHED.

The November 2 release-notes entry contains two parallel notices. Each one says the model is deprecated and no longer selectable in AI Studio, that requests will route to a named replacement on December 1, 2026, that billing will follow the replacement's rates, and that developers should change their model string and read a migration guide before that date. Google lists the breaking-change areas explicitly: function calling, `thought_signature` validation, `thinking_level` replacing `thinking_budget`, removed sampling parameters, and the end of prefilled model turns.

The same wording appears in caution callouts on the official pricing page, next to the Gemini 2.5 Pro and Gemini 2.5 Flash price lists. The deprecations page records the release dates of the two models (June 17, 2025 for both) and repeats that shutdown dates are the earliest possible retirement dates, with exact dates communicated in advance. In this case the routing date is already fixed.

The mechanism is the same one Google already uses inside the Gemini 3 family, where `gemini-3.7-flash` and `gemini-3.5-flash` keep their endpoints but are served by `gemini-3.8-flash` and `gemini-3.6-flash`. Existing code will keep receiving responses after December 1, but those responses will come from a different model with different defaults, different accepted parameters and a different price list. The notices are published in the Gemini Developer API documentation; they do not describe the model lifecycle on Vertex AI, which Google documents separately.

··········

GEMINI 2.5 PRO VS GEMINI 3.8 FLASH API PRICES BEFORE AND AFTER.

Prices below are as listed on the official Gemini Developer API pricing page on November 3, 2026, in USD per 1M tokens, paid tier. Gemini 2.5 Pro has two price bands, for prompts up to 200,000 tokens and above; Gemini 3.8 Flash has a single band for all prompt lengths and all input types, with a price change scheduled on January 1, 2027.

........

Item	Gemini 2.5 Pro (until Nov 30)	Gemini 3.8 Flash (Dec 1 to Dec 31)	Gemini 3.8 Flash (from Jan 1, 2027)
Input, prompts up to 200k	$1.25	$0.75	$1.50
Input, prompts over 200k	$2.50	$0.75	$1.50
Output incl. thinking, up to 200k	$10.00	$3.75	$7.50
Output incl. thinking, over 200k	$15.00	$3.75	$7.50
Context caching	$0.125 / $0.25	$0.075	$0.15
Cache storage, per 1M tokens per hour	$4.50	$0.50	$1.00
Batch input	$0.625 / $1.25	$0.375	$0.75
Batch output	$5.00 / $7.50	$1.875	$3.75
Priority input	$2.25 / $4.50	$1.35	$2.70
Priority output	$18.00 / $27.00	$6.75	$13.50

........

Where Gemini 2.5 Pro shows two values, the first applies to prompts up to 200,000 tokens and the second to longer prompts. Flex pricing for both models matches their Batch prices. The pricing page does not describe the Gemini 3.8 Flash rates before January 1 as a promotion; it simply lists two dated prices.

··········

GEMINI 2.5 FLASH VS GEMINI 3.5 FLASH-LITE API PRICES BEFORE AND AFTER.

The Gemini 2.5 Flash route lands on a model with an identical price for text, image and video, and a lower price for audio, because Gemini 3.5 Flash-Lite charges one input rate for every modality.

........

Item	Gemini 2.5 Flash (until Nov 30)	Gemini 3.5 Flash-Lite (from Dec 1)
Input, text / image / video	$0.30	$0.30
Input, audio	$1.00	$0.30
Output incl. thinking	$2.50	$2.50
Context caching, text / image / video	$0.03	$0.03
Context caching, audio	$0.10	$0.03
Cache storage, per 1M tokens per hour	$1.00	$1.00
Batch input, text / image / video	$0.15	$0.15
Batch input, audio	$0.50	$0.15
Batch output	$1.25	$1.25
Batch context caching	$0.03	$0.015
Priority input / output	$0.54 / $4.50	$0.54 / $4.50

........

Both models have a free tier on the Gemini Developer API with input and output listed as free of charge; context caching is not available on the Gemini 3.5 Flash-Lite free tier. Gemini 3.5 Flash-Lite has a 1M-token context window with up to 64,000 output tokens, and Google positions it as the fastest and lowest-cost model of the 3.5 family.

··········

[[ADPLUS_300x250_1]]

WHAT REROUTED GEMINI 2.5 WORKLOADS WILL COST IN DECEMBER AND JANUARY.

The figures below apply the official list prices to three illustrative workloads. They assume the new model produces the same number of output tokens as the old one, which will not hold exactly, since thinking tokens are billed as output and the newer models use different thinking defaults. Cache storage, grounding and tool fees are excluded.

A customer-support assistant on Gemini 2.5 Pro handling 20,000 requests a day, with 6,000 input tokens and 800 output tokens per request, consumes 3.6 billion input tokens and 480 million output tokens over a 30-day month. At Gemini 2.5 Pro rates that is 3,600 × $1.25 plus 480 × $10.00, or $9,300. The same traffic billed as Gemini 3.8 Flash in December costs 3,600 × $0.75 plus 480 × $3.75, or $4,500, and from January 2027 it costs 3,600 × $1.50 plus 480 × $7.50, or $9,000. If half of the input is served from the context cache, the monthly figure is $7,275 on Gemini 2.5 Pro, $3,285 on Gemini 3.8 Flash in December and $6,570 from January.

A long-document workload that sends 1,000 prompts of 300,000 tokens a month and receives 4,000 output tokens each falls into the more expensive Gemini 2.5 Pro band: 300 × $2.50 plus 4 × $15.00 gives $810. Because Gemini 3.8 Flash has no long-context surcharge, the same month costs $240 in December and $480 from January.

A Gemini 2.5 Flash classification pipeline running 1 million requests a month at 1,500 input and 200 output tokens costs 1,500 × $0.30 plus 200 × $2.50, or $950, before and after the switch to Gemini 3.5 Flash-Lite. An audio pipeline sending 100 million audio input tokens a month sees that line fall from $100 to $30.

........

Workload (30 days)	Before	December 2026	From January 2027
Support assistant, 2.5 Pro to 3.8 Flash	$9,300	$4,500	$9,000
Same, 50% cached input	$7,275	$3,285	$6,570
Long documents over 200k, 2.5 Pro to 3.8 Flash	$810	$240	$480
Classification, 2.5 Flash to 3.5 Flash-Lite	$950	$950	$950
Audio input, 100M tokens, 2.5 Flash to 3.5 Flash-Lite	$100	$30	$30

........

The change favours Gemini 2.5 Pro users with long prompts, output-heavy jobs and batch traffic most, since those are the items where Gemini 3.8 Flash is cheapest even after January. Input-heavy chat workloads on short prompts end up close to their old bill from January, with input costing more and output less. Gemini 2.5 Flash users pay the same for text and less for audio.

··········

GEMINI 3.8 FLASH AND 3.5 FLASH-LITE MIGRATION CHANGES FOR DEVELOPERS.

The routing keeps endpoints alive, but requests written for Gemini 2.5 can fail on Gemini 3.x models. The migration guides list concrete differences, and the safest course is to switch the model string before December 1 and test, so that the change does not happen silently in production.

Thinking control changes first. Gemini 3.x models use `thinking_level` instead of `thinking_budget`. Gemini 3.8 Flash accepts the levels low, medium and high, and its model page states that `minimal` returns an error. Gemini 3.5 Flash-Lite defaults to `minimal`; Google recommends keeping it for extraction, routing and classification, and moving to `medium` or `high` for subagents, tool calls, code execution and multi-step reasoning, to avoid premature tool termination.

Sampling parameters are also on the way out. The guides tell developers to remove `temperature`, `top_p` and `top_k`, which are deprecated and ignored today and will return HTTP 400 in future generations, and to remove `candidate_count`, which Gemini 3 and later do not support. A request that ends with a non-empty `model` turn, the prefill technique some developers use to force an output format, returns HTTP 400; Google points to `system_instruction` or structured outputs instead.

Function calling needs attention as well. Thought signatures must be preserved as Gemini 3.x requires, with an updated SDK, multimodal assets go inside the function response payload, and Google documents workarounds for `Malformed_Function_Call` errors linked to text emitted before a tool call.

The replacement models also differ in capability. Gemini 3.8 Flash, released on September 2, 2026, accepts text, image, video, audio and PDF with a 1,048,576-token input limit and 65,536 output tokens, outputs text only, and supports Batch, caching, code execution, computer use, file search, function calling, Search and Maps grounding, structured outputs and URL context. It does not support the Live API. Our guide to Gemini API pricing in 2026 covers the full price list of the current models, and our earlier report on Gemini 3.8 Flash pricing, coding and agents describes the model's launch. Teams reviewing their Gemini setup this month may also want to check the recent change that turned on request storage by default for new paid projects.

··········

GEMINI 2.5 RETIREMENT ON DECEMBER 1: CONFIRMED FACTS AND OPEN POINTS.

Google's documentation confirms the date, the two replacement models, billing at the replacement's rates, the end of AI Studio selection, the restriction of API access to past users, and the list of breaking-change areas. The official pricing page confirms every rate quoted here, including the scheduled January 1, 2027 increase for Gemini 3.8 Flash. Gemini 2.5 Flash-Lite stays available with no shutdown date.

Several points are not addressed in the published notices. Google does not say whether responses will indicate that a rerouted request was served by the replacement model, does not describe what happens to explicit context caches created on Gemini 2.5 models, and does not state how the change applies to Vertex AI. The real cost of each rerouted workload will also depend on how many thinking tokens the new models generate for the same prompts, which only a test on production traffic can measure. For Gemini 2.5 Pro customers the decisive dates are December 1, when the switch happens at lower prices, and January 1, when Gemini 3.8 Flash input becomes more expensive than the model it replaced.

··········

·····

DATA STUDIOS

·····

datastudios.org