LLMSwaps

Which model each app runs today, and when it last changed

Three identifiers ending together, none replaced

Three Veo identifiers share one shutdown date on the deprecations table, and two preview identifiers share an earlier one. No row names a successor. The newer generation exists and is not presented anywhere on that page as where the retired identifiers go. As of 2026-09-22.

What has been announced, entry by entryFilled where the vendor has announced something in that column, hollow where it has not.What has been announced, entry by entryWhat Gemini API publishesWhat Gemini APIpublishesWhat that leaves openWhat that leavesopenIs a successor named per rowIs a successor named per row — What Gemini API publishes: No replacement column appears on the tableIs a successor named per row — What that leaves open: The table carries identifiers and dates onlyHow many video lines share a dateHow many video lines share a date — What Gemini API publishes: Three on one date, two on an earlier oneHow many video lines share a date — What that leaves open: So the move is not staggered by versionIs the newer generation offered as the rou…Is the newer generation offered as the routeIs the newer generation offered as the route — What Gemini API publishes: Not stated on the deprecations pageIs the newer generation offered as the route — What that leaves open: It is documented elsewhere, as a productWhat the date itself commits toWhat the date itself commits to — What Gemini API publishes: The earliest a model might be retiredWhat the date itself commits to — What that leaves open: A floor, so the runway is not fixed either
Fig. 1 Filled where the vendor has announced something in that column, hollow where it has not.
What the deprecations table says about the route off each retired identifier. Recorded 2026-09-22.
What the page is askedWhat Gemini API publishesWhat that leaves open
Is a successor named per rowNo replacement column appears on the tableThe table carries identifiers and dates only
How many video lines share a dateThree on one date, two on an earlier oneSo the move is not staggered by version
Is the newer generation offered as the routeNot stated on the deprecations pageIt is documented elsewhere, as a product
What the date itself commits toThe earliest a model might be retiredA floor, so the runway is not fixed either

Inclusion rule. Read from the deprecations page for this interface, restricted to video identifiers. Product documentation for newer models is not read into this column. Order. What the table publishes first, then what it does not.

1Retiring three lines on one date is a decision about people

Staggering retirements spreads the work of migrating; collapsing them into one date concentrates it. Three identifiers going at once means every caller of any of them has the same deadline and the same week to test.

That is defensible for a platform with many customers and no successor to offer per line. It also means the register cannot describe a gradual path off this generation, because the page does not describe one.

2A table with no successor column

The shape of the page decides what can be published. This one has identifiers and dates and no place to put a destination, so the absence is structural rather than an omission in a particular row.

Compare the platform in this column that does name one: it publishes a schedule with a replacement written into the row itself. Same kind of document, one extra column, and the difference in planning value is large.

The page shape decides what can be publishedA retirement schedule with a replacement column can name a successor per row. A deprecations table with identifiers and dates has nowhere to put one, so the absence here is structural rather than an omission in a particular row.Identifiers and datesSchedule with a successorColumnsModel, shutdown dateModel, date, replacementAnswers whenYes, as an earliest dateYes, as a fixed dateAnswers where toNowhere on the pageIn the row itselfUsed byThis entryOne platform in the registerTwo lifecycle pages, two different questions answered
Fig. 2 Both documents are lifecycle pages; one extra column is the whole difference in planning value.

3Where the successor information actually lives

Newer identifiers are documented as products, on pages written to sell capability rather than to manage an ending. A reader can find them, and nothing connects them to the rows that are going away.

For anyone reaching these weights through an application the gap is wider still. The reseller repeats neither the dates nor the newer identifiers, so the entire migration question exists two layers above the person whose series depends on it.

  • veo-3.0-generate-001
    30 June 2026veo-3.0-fast-generate-001 and veo-2.0-generate-001, shutdown dateGoogle, Gemini deprecations / recorded 2026-09-22
  • veo-3.0-generate-preview and veo-3.0-fast-generate-preview
    12 November 2025shutdown dateGoogle, Gemini deprecations / recorded 2026-09-22
  • What a date in that table commits to
    The shutdown dates listed indicate the earliest possible dates on which a model might be retiredGoogle, Gemini deprecations / recorded 2026-09-22

4Sources

Read from the deprecations page on ai.google.dev, 2026-09-22. The whole entry is on Gemini API, and the column it sits in is on Named replacement. Same column, other entries: Amazon Bedrock, LTX Studio.