Google schedules gemini-2.5-flash-image for shutdown on October 2, 2026, and names gemini-3.1-flash-image-preview as the replacement

Google schedules gemini-2.5-flash-image for shutdown on October 2, 2026, and names gemini-3.1-flash-image-preview as the replacement | Narracomm Intelligence Record IDN/A DateOctober 1, 2026 BeatAI model lifecycle JurisdictionN/A EntityGoogle Gemini API (gemini-2.5-flash-image) Instrument / citegemini-2.5-flash-image shutdown schedule, October 2, 2026. Source: Gemini API deprecations, Google AI for Developers

The named replacement already carries an earlier shutdown date of June 25, 2026 on the same page, and that row points on to gemini-3.1-flash-image.

In brief: Google’s Gemini API deprecations page lists gemini-2.5-flash-image for shutdown on October 2, 2026, with gemini-3.1-flash-image-preview as the recommended replacement.

Google has set October 2, 2026, as the shutdown date for gemini-2.5-flash-image and has named gemini-3.1-flash-image-preview as the recommended replacement. That replacement is not an open-ended target on the same record. Its own row lists a shutdown date of June 25, 2026, and points onward to gemini-3.1-flash-image. A team that follows the replacement cell literally lands on a model id the page has already dated for an earlier shutdown.

The controlling page is the Gemini API deprecations document on Google AI for Developers, last updated 2026-10-01 UTC. It “lists the known deprecation schedules for stable (GA) and preview models and for managed agents in the Gemini API.” Under Gemini 2.5 Flash models, the row for gemini-2.5-flash-image gives a release date of October 2, 2025, a shutdown date of October 2, 2026, and a recommended replacement of gemini-3.1-flash-image-preview. The id sits with the stable entries in that section, not under the “Preview models” subheading. The scheduled life on this page is one year to the day.

The page separates announcement from cutoff. “A deprecation is the announcement that we no longer provide support for a model, and that it will be shut down in the near future.” Shutdown is terminal: “Once a model is shutdown, it is completely turned off, and the endpoint is no longer available.” Announcements, the page says, are made on the Release notes page; “the announced earliest shutdown dates are tracked on this page.” The October 2, 2026 figure is therefore an earliest shutdown date, not a statement that the endpoint has already gone dark. “Already-shutdown models are indicated with gray backgrounds.” This reading of the table does not attach that marker to the gemini-2.5-flash-image row beyond the date columns themselves.

The replacement cell does not point at the model with no announced shutdown. In the Gemini 3 models section, gemini-3.1-flash-image-preview shows a release date of February 26, 2026, a shutdown date of June 25, 2026, and a recommended replacement of gemini-3.1-flash-image. gemini-3.1-flash-image shows a release date of May 28, 2026, and “No shutdown date announced.” gemini-3-pro-image, released the same day, also has no shutdown date announced. Its preview, gemini-3-pro-image-preview, carries the same June 25, 2026 shutdown date and points to gemini-3-pro-image. The page does not say why the 2.5 image row names the preview rather than the model that preview’s own row names.

June 25, 2026, precedes October 2, 2026. The page does not say, in the rows extracted here, that gemini-3.1-flash-image-preview has been turned off, and it does not reprint a gray-background flag in the text of the table. What it does publish is a shutdown date for the named replacement that falls before the shutdown date of the model being replaced. A migration written to the replacement cell on October 2 would be aimed at an id whose earliest shutdown date, on this page, has already passed.

The prior hop on the same table runs the other direction. gemini-2.5-flash-image-preview, listed under Preview models, has a release date of May 7, 2025, a shutdown date of January 15, 2026, and a recommended replacement of gemini-2.5-flash-image. The published path is preview to the stable image id, then from that stable id to a different preview. The page does not describe this as a sequence, and it does not say the January 15 date was executed. It only lists the dates and the replacement cells.

Sibling rows in the Gemini 2.5 Flash models table do not share the cutoff. gemini-2.5-flash, released June 17, 2025, has no shutdown date announced. gemini-2.5-flash-lite, released July 22, 2025, has none either. The image id is the row in that block with a dated shutdown. The page does not give a reason for the difference, and it does not attach the general line about retiring models to this entry. There is no notice-period floor on the page, no email-notification rule, and no dedicated-capacity clause.

What an enterprise buyer can verify from the primary record is the schedule and the pointer. gemini-2.5-flash-image is listed for shutdown on October 2, 2026, one year after its listed release date, and the recommended replacement is gemini-3.1-flash-image-preview. What the record does not settle is whether that preview endpoint is still available, why it is the named replacement when its own row points to gemini-3.1-flash-image, or whether a workload that needs an image model with no announced shutdown date should pin gemini-3.1-flash-image instead. The page tracks earliest shutdown dates. It does not confirm that the October 2 cutoff has occurred.

Why it matters

A founder who tells a customer or an auditor that the migration target is the model in the replacement cell is citing a pointer to an id the same table has already dated for shutdown. The image-generation path on this API is a chain, not a single hop, and the model at the end of that chain with no shutdown date announced — gemini-3.1-flash-image — is not the model the gemini-2.5-flash-image row names. That gap is the diligence item. The page does not close it.

Sources

Subscribe — you own it

No tracking, no middleman. Follow by RSS (nothing is collected) — or add your email to our self-hosted list.

RSS feed →
Subscribe
Notify of
0 Comments
Oldest
Newest Most Voted
0
Would love your thoughts, please comment.x
()
x