Gemini 2.5 Pro and Flash Deprecated: What Changes on December 1?

November 3, 2026 · AI Radar

If your automation says gemini-2.5-flash somewhere, this post is for you. On November 2, Google added one line to the Gemini API changelog: 2.5 Pro and 2.5 Flash are now deprecated. Starting December 1, 2026, requests sent under those names won't return an error; they'll quietly go to a different model.

The "quietly" part matters. Your code keeps running and you see no error on screen, but a different model writes the answer, and you're billed at that model's price. Finding this out now is cheaper than noticing it on your end-of-month bill.

What changed?

The Gemini API changelog entry dated November 2, 2026 says this: gemini-2.5-pro and gemini-2.5-flash are deprecated; starting December 1, Pro requests go to gemini-3.8-flash and Flash requests go to gemini-3.5-flash-lite, and model names need to be updated before that date [A]. The deprecations page gives the details: both models have already been removed from Google AI Studio, API access is limited to accounts actively using the model, and the shutdown date listed is "December 1, 2026 at the earliest" [A]. In other words, the date can move later, not earlier. For redirected requests, you'll need to check the migration guide, which covers the new model's breaking API changes [A].

Another of Google's doors, Vertex AI, moved earlier. Vertex's model versions page lists October 20, 2026 as the retirement date for both models; the recommended replacements are 3.8 Flash or 3.5 Flash-Lite [A]. The date on the Gemini API is December 1, on Vertex it's October 20; the same model has two different timelines. Which one affects you depends on which door you use.

There's also something that isn't shutting down. gemini-2.5-flash-lite has not been deprecated; the page says it remains available "until further notice" and lists no replacement model [A].

Source: Gemini API pricing page and deprecations page, as of November 3, 2026.

What will it cost?

According to the Gemini API pricing page, here are the numbers (per million tokens, US dollars, text) [A]:

Pricing page as of November 3, 2026 [A]. In the Flash and Flash-Lite rows, audio input is priced separately; the figures above are for text, image and video.

Two things stand out here. First, moving from 2.5 Flash to 3.5 Flash-Lite doesn't change the price: both list at 0.30 / 2.50. The bill stays the same; what changes is the model itself. Second, moving from Pro to 3.8 Flash is cheap in December, because 3.8 Flash's discounted price ends on December 31. After January 1, the input side is even more expensive than Pro's old price (1.50 versus 1.25), while the output side is still cheaper (7.50 versus 10). If you feed it very long documents and get short answers, your bill may go up; if you ask short questions and get long answers, it goes down. Work out which one your workload looks like and calculate accordingly.

Whether the redirected model's answer quality will match the old one isn't clear in the official source. A job running under the Pro name will be running on a Flash model after December 1. In my view, that part matters more than the price difference.

Who does this affect?

• People building flows that connect to Gemini through n8n, Make or Zapier: Search for the gemini-2.5-... names you typed by hand into an HTTP node or a model field. Nobody is going to change a hand-typed model name for you. Tool pages: n8n(in Turkish), Make(in Turkish), Zapier(in Turkish).
• Agencies and developers using the Gemini API for clients: Telling the client "the model changed, the output may be a bit different" is better than the client noticing it first.
• People who only use the Gemini app: This announcement is about the API and doesn't directly affect you. Model selection in the app is a separate topic; see the Gemini 4 Argon post.
• People working through Vertex AI: The October 20 retirement date has passed. If you're seeing errors or redirects in production, this may be the reason.

Before December 1

0/5

• I listed which flows have an old Gemini model name written in them
• I tested whether the job that uses Pro holds up in quality when it drops to Flash
• I calculated separate budgets for December and January (3.8 Flash doubles on January 1)
• I told the client or the team about the model change
• If I use Vertex AI, I checked how the October 20 retirement affects my account

Check off the items; the empty ones are where to look before November 30.

AI Radar

Frequently asked questions

When exactly are Gemini 2.5 Pro and Flash shutting down?

For the Gemini API, the earliest date is December 1, 2026; from that day, requests under the old names are redirected to the new model. Google gives December 1 as the "earliest" date and says it will announce the exact day in advance. On Vertex AI, the retirement date is listed as October 20, 2026. As of November 3, 2026, the models have been removed from the Google AI Studio interface.

If I use the Gemini app, do I need to do anything?

No. The announcement concerns people calling the API by model name. The official source doesn't mention any change to model selection in the app.

If I don't change the old model name, will my workflow stop?

On the Gemini API it won't stop; it gets redirected. The real risk is that the answer comes from a different model and you're billed at the new model's price. You'll need to read about the redirected model's breaking changes in Google's migration guide.

If you'd like to talk through which model makes sense for which job in your business, see the AI consulting page; if you'd like to update your Gemini API flows together, see the AI automations page. To see where you stand first: START YOUR FREE ANALYSIS .

Sources

• [A] Gemini API changelog, November 2, 2026 entry
• [A] Gemini API: model deprecations page, as of November 3, 2026
• [A] Gemini API pricing page, as of November 3, 2026
• [A] Vertex AI: model versions and retirement dates, as of November 3, 2026