Back to HomeAI API

Gemini 2.5 Pro and Flash Auto-Route From December 1: Which Model You Get, Pricing, and the Code to Change (2026)

15 min min read
#Gemini 2.5 Pro#Gemini 2.5 Flash#Gemini 3.8 Flash#Gemini 3.5 Flash-Lite#Gemini API#Google#AI API#API Pricing#Model Migration#Model Deprecation#Enterprise Procurement

Illustration of API requests to older Gemini Pro and Flash models being automatically redirected to newer Flash and Flash-Lite models after the routing date: two request streams pass through a railway switch and change tracks, while the bill beside them switches to the new models' price list

What Happened: Announced November 2, Routing From December 1

The official Gemini API changelog posted two deprecation notices on 2026-11-02. gemini-2.5-pro and gemini-2.5-flash, both stable models released on 2025-06-17, are now marked deprecated and "no longer available to select in Google AI Studio."

The sentence that matters for your bill and your code comes next: starting 2026-12-01, all API requests to these two models will automatically route to newer models and be billed at the newer models' rates. Your code needs to be updated before 2026-12-01.

Current modelReleasedRoutes to from 2026-12-01Billing
gemini-2.5-pro2025-06-17gemini-3.8-flashGemini 3.8 Flash rates
gemini-2.5-flash2025-06-17gemini-3.5-flash-liteGemini 3.5 Flash-Lite rates
gemini-2.5-flash-lite2025-07-22Not routed; no shutdown date announcedUnchanged

Sources: the 2026-11-02 entry of the Gemini API changelog and the Gemini API deprecations page, checked 2026-11-19.

How Routing Differs from Shutdown

The deprecations page defines a shutdown as the model being completely turned off, with the endpoint no longer available. That is not what happens to 2.5 Pro/Flash: the endpoints keep accepting requests, but a newer model handles them. That means:

  1. Requests won't fail because of the model name: code that still sends gemini-2.5-pro will still get responses after 12-01
  2. But the response comes from a different model: 3.8 Flash answers for 2.5 Pro and 3.5 Flash-Lite answers for 2.5 Flash, so output quality, style, and token usage may all differ
  3. The bill switches price lists: the same code at the same traffic is billed at the new model's prices in December
  4. Breaking changes apply: Google lists the items in the next section as breaking API changes, so code you haven't updated may error out or behave differently

Note: this differs from OpenAI. OpenAI's deprecations page says a model stops accepting requests on its shutdown date and does not mention automatic routing; see the OpenAI model shutdown schedule.

Deprecations-Page Dates Are the "Earliest Possible" Dates

Above its tables, the Gemini API deprecations page states that the listed shutdown dates are the earliest possible dates on which a model might be retired, and that exact dates will be communicated in advance. The 2.5 Pro/Flash routing, however, is stated in both the changelog and the pricing page as "Effective December 1, 2026," so treat 12-01 as your deadline.

The other 2.5-family dates on the same page are also earliest possible dates:

ModelDate on the deprecations pageOfficial replacement
gemini-2.5-flash-preview-tts, gemini-2.5-pro-preview-tts2026-12-18gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts
gemini-2.5-flash-image2027-03-15gemini-3.1-flash-lite-image
gemini-2.5-flash-native-audio-preview-12-20252027-03-31gemini-3.8-live

The deprecations page does not say these will be routed automatically.

Prices Before and After Routing

Item (per million tokens, Standard paid tier)Gemini 2.5 Pro (before)Gemini 3.8 Flash (after)
Input$1.25 (prompts ≤200K tokens); $2.50 (>200K)$0.75 through 2026-12-31; $1.50 from 2027-01-01
Output (including thinking)$10 (≤200K); $15 (>200K)$3.75 through 2026-12-31; $7.50 from 2027-01-01
Context caching$0.125 (≤200K); $0.25 (>200K)$0.075 through 2026-12-31; $0.15 from 2027-01-01
Cache storage (per million tokens per hour)$4.50$0.50 through 2026-12-31; $1 from 2027-01-01
Tiered by prompt lengthYes, split at 200K tokensNo tiers on the pricing page
Free tierYesYes
Item (per million tokens, Standard paid tier)Gemini 2.5 Flash (before)Gemini 3.5 Flash-Lite (after)
Input (text, image, video)$0.30$0.30
Input (audio)$1.00$0.30 (same price for all input types)
Output (including thinking)$2.50$2.50
Context caching$0.03 (audio $0.10)$0.03
Cache storage (per million tokens per hour)$1.00$1.00
Free tierYesYes (no caching on the free tier)

Source: Gemini Developer API pricing, checked 2026-11-19, USD. Batch input and output prices are half of Standard.

A few things stand out:

  • 2.5 Pro to 3.8 Flash: in December 2026, both input and output unit prices are lower than 2.5 Pro's. But 3.8 Flash's promotional price only runs through 2026-12-31; from 2027-01-01 its $1.50 input is higher than 2.5 Pro's $1.25, while its $7.50 output is still below $10
  • 2.5 Pro's long-prompt surcharge goes away: 2.5 Pro charges $2.50/$15 for prompts over 200K tokens; 3.8 Flash's pricing page has no prompt-length tiers
  • 2.5 Flash to 3.5 Flash-Lite: text prices are identical, so the bill changes mainly for audio input ($1.00 down to $0.30); the real question is whether quality holds up after moving from a Flash model to a Flash-Lite model

What the Two Routing Targets Are

  • Gemini 3.8 Flash (gemini-3.8-flash): GA on 2026-09-02; the changelog calls it "our most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, and complex enterprise workflows." Google now labels it a previous-generation Flash model
  • Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite): released 2026-07-21; the changelog describes it as "a low-latency, highly cost-effective subagent option designed for high-volume automation." Google recommends Gemini 4 Hydrogen or Gemini 3.5 Flash-Lite for new projects

Neither has a shutdown date announced on the deprecations page.

Cost Math (Arithmetic, Not Benchmarks)

Assumptions: Standard paid tier, no caching or Batch; token counts are treated as unchanged across the switch (the official pages don't say how token counts change between models, and thinking tokens are billed as output, so measure real usage after go-live).

Scenario A: General 2.5 Pro Work, 50M Input and 10M Output per Month, All Prompts Under 200K Tokens

CalculationInput costOutput costMonthly total
Gemini 2.5 Pro (before)50M × $1.25 = $62.5010M × $10 = $100$162.50
Routed to 3.8 Flash (December 2026)50M × $0.75 = $37.5010M × $3.75 = $37.50$75
Routed to 3.8 Flash (from 2027-01-01)50M × $1.50 = $7510M × $7.50 = $75$150
(Comparison) Gemini 4 Hydrogen50M × $1.50 = $7510M × $7.50 = $75$150

December's bill is about 46% of the original, but from 2027 it returns to about 92%. Budget with the prices from 2027-01-01; December's low price is only a bridge. From 2027, 3.8 Flash and Gemini 4 Hydrogen have the same unit prices.

Scenario B: Long 2.5 Pro Documents, 1,000 Requests a Month at 300K Input and 3,000 Output Each

That totals 300M input and 3M output tokens a month. With prompts over 200K tokens, 2.5 Pro bills the whole request at the higher tier.

CalculationInput costOutput costMonthly total
Gemini 2.5 Pro (>200K, before)300M × $2.50 = $7503M × $15 = $45$795
Routed to 3.8 Flash (December 2026)300M × $0.75 = $2253M × $3.75 = $11.25$236.25
Routed to 3.8 Flash (from 2027-01-01)300M × $1.50 = $4503M × $7.50 = $22.50$472.50

Long-prompt work saves the most on price, still about 59% of the original from 2027. This assumes 3.8 Flash's context window can hold prompts this long; the official pages cited here don't list a context limit, so check the model page.

Scenario C: High-Volume 2.5 Flash Work, 200M Input and 20M Output per Month

CalculationText inputOutputMonthly total
Gemini 2.5 Flash (before)200M × $0.30 = $6020M × $2.50 = $50$110
Routed to 3.5 Flash-Lite200M × $0.30 = $6020M × $2.50 = $50$110
Switch to 3.8 Flash yourself (December 2026)200M × $0.75 = $15020M × $3.75 = $75$225
Switch to 3.8 Flash yourself (from 2027-01-01)200M × $1.50 = $30020M × $7.50 = $150$450

If the input is audio: with 100M audio input and 5M output tokens a month, 2.5 Flash costs 100M × $1.00 + 5M × $2.50 = $112.50, and routed to 3.5 Flash-Lite it costs 100M × $0.30 + $12.50 = $42.50, about 38% of the original.

The point of Scenario C isn't savings: the text bill doesn't change after routing, but the model moves from Flash to Flash-Lite. If quality falls short and you want to stay at the Flash level, switching to 3.8 Flash costs about 2 times as much (December 2026) and about 4 times as much from 2027.

The Code to Change: Four Breaking Changes and a Pre-Launch Checklist

The changelog lists the same breaking API changes for both models. Google's advice is to change the model string yourself before 2026-12-01 and read the Gemini 3.8 Flash and Gemini 3.5 Flash-Lite migration guides first (linked from the changelog entries).

ItemCommon 2.5-era usageWhat to do after routing
Model stringgemini-2.5-pro, gemini-2.5-flashChange to gemini-3.8-flash, gemini-3.5-flash-lite (or whichever model you choose after evaluation)
Function calling and thought_signatureFunction calls and thought signatures not handled under the new rulesListed as a validation change; check multi-turn function-calling requests against the migration guide
Thinking settingsthinking_budget (a token budget)Use thinking_level instead
Sampling parametersSetting temperature, top_p, top_kRemove them; the changelog announced these three sampling parameters as deprecated on 2026-07-21
Prefilled model turnsEnding the conversation with model-role text for the model to continueRemove it and control output format another way

Pre-launch checklist:

  1. Inventory: search code and config for gemini-2.5-pro and gemini-2.5-flash; gemini-2.5-flash-lite is out of scope this time, so don't change it by mistake
  2. Decide what each call site should become: accept the official routing target, or switch to another model using "How to Choose" below
  3. Handle the four parameter changes above, then run a regression pass, especially on function calling and multi-turn conversations
  4. Re-measure tokens and quality: thinking tokens are billed as output and usage may change with the model; for work moving from 2.5 Flash to Flash-Lite, confirm quality is good enough
  5. Rebuild the budget: 3.8 Flash's $0.75/$3.75 only runs through 2026-12-31; budget with the $1.50/$7.50 that applies from 2027-01-01
  6. Schedule other 2.5 models separately: the 2.5 TTS previews, gemini-2.5-flash-image, and the native audio preview each have their own earliest possible date (see the table above)
  7. AI Studio testing: 2.5 Pro/Flash can no longer be selected in AI Studio, so retest prompts tuned there on the new models

CloudInsight Enterprise Gemini API Procurement | Local Invoices, No Foreign Credit Card Needed

This routing moves your bill to new models' rates, and 3.8 Flash's prices double from 2027. What enterprises get stuck on most is often not the technology but payment, invoicing, and budget control.

Contact CloudInsight for a Gemini API Enterprise Quote

We offer one-stop procurement for Claude, GPT, and Gemini, Taiwan uniform invoices, and technical support in Chinese.

How to Choose: By the Model You Use Today

Using todayIf you do nothingModels you can switch to yourself
gemini-2.5-proRouted to 3.8 Flash from 12-01, billed at $0.75/$3.75 ($1.50/$7.50 from 2027-01-01)3.8 Flash (official replacement); Gemini 4 Hydrogen at $1.50/$7.50 (Google's recommendation for new projects); for stronger reasoning, evaluate Gemini 4 Neon at $3/$15 or Carbon ($2/$10 through 2026-11-30, $4/$20 from 12-01)
gemini-2.5-flashRouted to 3.5 Flash-Lite from 12-01, billed at $0.30/$2.503.5 Flash-Lite (official replacement, same text price); if quality falls short, evaluate 3.8 Flash or Gemini 4 Hydrogen
gemini-2.5-flash-liteNot affected; no shutdown date announcedKeep using it; Google states that the 2.5 models are available only to users who have used them in the past, and recommends Gemini 4 Hydrogen or 3.5 Flash-Lite for new projects
2.5 TTS previews, 2.5 Flash Image, 2.5 native audio previewRetire on their earliest possible dates; no routing statedMove to the official replacements on the deprecations page

For positioning and pricing details on the three Gemini 4 models, see Gemini 4 Carbon, Neon, or Hydrogen?

FAQ: Gemini 2.5 Pro and Flash Routing

Can I still call Gemini 2.5 Pro after December 1?

Yes, but 2.5 Pro won't be the one answering. Starting 2026-12-01, requests to gemini-2.5-pro will automatically route to gemini-3.8-flash, and gemini-2.5-flash will route to gemini-3.5-flash-lite. Neither can be selected in AI Studio anymore.

How are routed requests billed?

At the new model's prices. Requests routed to 3.8 Flash cost $0.75/$3.75 per million input/output tokens through 2026-12-31 and $1.50/$7.50 from 2027-01-01; requests routed to 3.5 Flash-Lite cost $0.30/$2.50.

What happens if I don't change my code?

The model name won't make requests fail, but Google lists breaking changes: function calling and thought_signature validation, thinking_level replacing thinking_budget, and removed sampling parameters and prefilled model turns. Code that uses these may error out or behave differently, and Google recommends updating your model string and reading the migration guides before 12-01.

Will Gemini 2.5 Flash-Lite be routed too?

No. On the official deprecations page, gemini-2.5-flash-lite has no shutdown date announced and no replacement listed.

Are the dates on the deprecations page final?

The deprecations page says its dates are the earliest possible retirement dates and exact dates are announced in advance. The 2.5 Pro/Flash routing, however, is stated in both the changelog and the pricing page as effective 2026-12-01, so treat it as a deadline.

Will routing make things cheaper?

It depends on the model and the month. For 2.5 Pro routed to 3.8 Flash, December 2026 unit prices are lower, but from 2027-01-01 the $1.50 input price is higher than 2.5 Pro's $1.25. For 2.5 Flash routed to 3.5 Flash-Lite, text prices don't change and audio input gets cheaper.

Conclusion: Don't Wait for Routing, Switch Before December 1

Three Core Recommendations

  1. Change your model strings before December 1: routing hands your requests to a different model, and switching yourself lets you test quality and the breaking changes first
  2. Handle the four parameter changes first: replace thinking_budget with thinking_level, remove sampling parameters and prefilled turns, and check function calling against the migration guide
  3. Budget with 2027 prices: 3.8 Flash's $0.75/$3.75 only runs through 2026-12-31, then $1.50/$7.50

Ready to Plan Your Migration?

Contact the CloudInsight sales team and we'll help evaluate replacements and usage mix for 2.5 Pro/Flash and handle payment and invoicing.

Join our LINE Official Account for instant model-selection consultation.


Further Reading

References

Need Professional Cloud Advice?

Whether you're evaluating cloud platforms, optimizing existing architecture, or looking for cost-saving solutions, we can help

Book Free Consultation

Related Articles

AI API

Gemini 4 Argon Guide: API Pricing (Intro Rate Through Nov 30), Specs, and Migrating from Gemini 3

Google made Gemini 4 Argon generally available (GA) to developers on 2026-10-13; since 11-10 the strongest-reasoning-model slot belongs to Gemini 4 Carbon, and Argon remains available. The introductory price of $2/$10 only runs through 2026-11-30; from 12-01 it moves to $4/$20, and there is no free tier. This guide covers the official specs (1M context, output up to 1M, three thinking levels), both price stages for every consumption option, cost math, how it splits work with Gemini 3.8 Flash, and the settings to change when moving from the Gemini 3 family.

AI API

Gemini 4 Carbon, Neon, or Hydrogen? API Pricing, Free Tier, and How They Differ from Argon (2026)

Google took three Gemini 4 models to GA in November 2026: Carbon, its strongest reasoning model (11-10); Hydrogen, built for everyday efficiency (11-12); and Neon, built for fast task completion (11-18). Carbon's $2/$10 rate runs only through 2026-11-30 and becomes $4/$20 on 12-01; Neon is a flat $3/$15; Hydrogen is $1.50/$7.50 and the only Gemini 4 model with a free tier. This guide covers official positioning, prices across billing modes, thinking levels, cost math, and how to split work now that Argon has become the frontier model that spends fewer tokens than Carbon.

AI API

Claude Fable 5.2 Guide (2026): API Pricing, Differences from Fable 5.1, Mythos 5.2, and Migration

Anthropic released Claude Fable 5.2 on 2026-11-09 at $8/$40 per million input/output tokens, 20% less than Fable 5.1's $10/$50, with cache and batch prices lowered by the same proportion and the full 1M-token context at standard pricing. This guide covers official specs and prices, Mythos 5.2 (same specs and price, verified organizations only), cost math, and the changes that return 400 errors when you move over from Fable 5.1 or Opus 5.5.