As of November 3, 2026. On November 2, 2026, Google deprecated gemini-2.5-pro and gemini-2.5-flash in the Gemini API (Gemini API changelog). Starting December 1, 2026, requests sent to these model IDs are automatically routed to newer models and billed at the newer models' rates. Both models are already gone from the model picker in Google AI Studio. If any internal tool or vendor-built system calls 2.5 by name, test it against the replacement model during November.
The short answer: 2.5 Pro routes to 3.8 Flash, 2.5 Flash routes to 3.5 Flash-Lite
Model ID you call today | Routed to from December 1 | Billed at | Official migration guide |
|---|---|---|---|
|
| Gemini 3.8 Flash rates | What's new in Gemini 3.8 Flash |
|
| Gemini 3.5 Flash-Lite rates | What's new in Gemini 3.6 (3.5 Flash-Lite section) |
Automatic routing does not mean you can ignore this. Google's changelog explicitly says the move includes breaking API changes, so some requests written for 2.5 will fail on the replacement model. The five changes that matter most are listed below.
gemini-2.5-flash-lite is not part of this announcement. As of November 3, no deprecation has been announced for it.
The Deprecations page has not caught up yet
As of November 3, the official Deprecations page still lists both 2.5 Pro and 2.5 Flash as "No shutdown date announced," along with the older note that 2.5 access is limited to existing users. The changelog entry dated November 2 is the newer source, so use it as the basis for planning.
What happens to your bill
Paid Tier standard prices per 1M tokens, from the pricing page:
Model | Input | Output (incl. thinking) |
|---|---|---|
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 Dec 31, 2026) | $0.75 | $3.75 |
Gemini 3.8 Flash (from Jan 1, 2027) | $1.50 | $7.50 |
Gemini 2.5 Flash | $0.30 (audio $1.00) | $2.50 |
Gemini 3.5 Flash-Lite | $0.30 (text, image, video and audio) | $2.50 |
Worked example: 10M input and 2M output tokens per month
- Gemini 2.5 Pro: $12.50 input + $20.00 output = $32.50/month
- Gemini 3.8 Flash (December): $7.50 + $7.50 = $15.00/month
- Gemini 3.8 Flash (from January 2027): $15.00 + $15.00 = $30.00/month
The formula is input tokens ÷ 1M × input price + output tokens ÷ 1M × output price. This assumes the same token counts. Google's migration guide notes that 3.8 Flash "can use more tokens on longer running and complex tasks, by design," and its default thinking_level is medium. If output tokens grow, 3.8 Flash can end up costing more than 2.5 Pro after the January price change, so measure real token usage during testing.
Moving from 2.5 Flash to 3.5 Flash-Lite keeps the same text pricing, and audio input drops from $1.00 to $0.30.
Grounding with Google Search is counted differently
Model | Free allowance | After that |
|---|---|---|
Gemini 2.5 Pro / 2.5 Flash | 1,500 requests per day | $35 per 1,000 grounded prompts |
Gemini 3.8 Flash / 3.5 Flash-Lite (shared across Gemini 3 and newer) | 5,000 per month | $14 per 1,000 search queries |
The free allowance changes from 1,500 per day to 5,000 per month, and billing moves from "per grounded prompt" to "per search query." The pricing page notes that a single request can trigger more than one search query, and each one is charged. If you ground hundreds of requests a day, check your volume before December.
Five things that break 2.5-era code
From the two migration guides (Gemini 3.8 Flash and Gemini 3.6 / 3.5 Flash-Lite):
Change | What happens | Fix |
|---|---|---|
| 2.5's token budget for thinking no longer applies | Use the string enum |
| Deprecated and ignored today; future model generations return HTTP 400 | Remove them and control consistency with system instructions |
| Unsupported in Gemini 3 and later | Send separate requests if you need multiple drafts |
Prefilled model turns | A request whose last non-empty turn is | End on a user turn; keep multi-turn history server-side with |
Function calling | With the generateContent API, every | Add both fields and use an SDK that preserves thought signatures |
The most common trap is the 2.5-era trick of prefilling the start of the model's answer (for example, an opening brace to force JSON). After December 1, that request returns a 400 on the replacement model.
thinking_level values differ by model
Gemini 3.8 Flash does not support minimal and returns an error if you set it; use low, medium or high. Gemini 3.5 Flash-Lite defaults to minimal. Google recommends leaving it there for high-volume extraction, routing and classification, and raising it to medium or high for agents that call tools or reason over several steps, otherwise they may stop tool use too early.
Migration checklist before December 1
- Find every call site. Search code, no-code tools and vendor systems for
gemini-2.5-proandgemini-2.5-flash, including environment variables and settings screens. - Pick the target. Accept the automatic target, or choose by workload: 3.8 Flash for heavy reasoning and long tasks, 3.5 Flash-Lite for cheap high-volume work, or Gemini 3.6 Flash in between (Google says it remains supported).
- Fix the five items above.
- Replay real inputs in staging. Run a few dozen representative requests and compare quality and token counts against 2.5.
- Switch the model ID yourself. Don't rely on silent routing; an explicit change keeps your audit trail clear.
- Budget for January. Gemini 3.8 Flash input and output prices double on January 1, 2027.
If you are also creating a new project for the migration, note that Generate Content API requests in paid projects created on or after October 26, 2026 are stored by default. See how to turn off Gemini API logging.
Model retirements now arrive every year for teams that call LLM APIs directly. At Mihata, we build custom AI tools for small businesses with the model ID in one place and a test routine for migrations like this one. If you'd like a second pair of hands on your migration, we'd be glad if you took a look.
Summary
Gemini 2.5 Pro and 2.5 Flash were deprecated on November 2, 2026. From December 1, they route to Gemini 3.8 Flash and Gemini 3.5 Flash-Lite and are billed at those rates. Because 2.5-style parameters such as thinking_budget and prefilled model turns can fail after routing, change the model ID yourself in November, test quality and token usage, and plan for the January 2027 price increase on 3.8 Flash.
FAQ
When does Gemini 2.5 Pro stop working?
Google deprecated gemini-2.5-pro on November 2, 2026 and removed it from the Google AI Studio model picker. From December 1, 2026, API requests to gemini-2.5-pro are automatically routed to gemini-3.8-flash and billed at Gemini 3.8 Flash rates.
What replaces Gemini 2.5 Flash?
From December 1, 2026, requests to gemini-2.5-flash are routed to gemini-3.5-flash-lite and billed at Gemini 3.5 Flash-Lite rates. Text pricing stays at $0.30 input and $2.50 output per 1M tokens. As of November 3, gemini-2.5-flash-lite has no announced deprecation.
If routing is automatic, do I need to do anything?
Yes. Google lists breaking API changes, including thinking_budget being replaced by thinking_level and prefilled model turns returning HTTP 400. Update the model ID yourself and test before December 1.
Will my Gemini bill go up?
Moving from 2.5 Pro to 3.8 Flash is cheaper per token until December 31, 2026 ($0.75/$3.75 vs $1.25/$10.00). From January 1, 2027, 3.8 Flash costs $1.50/$7.50, and it can use more tokens by design, so recalculate with real usage.