LLMSwaps

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

Which model each app runs, and when it last changed

An app notice records one thing precisely: which model the product tells its users it runs on, and when that statement last changed. Everything else about the app belongs on the app's own site. As of 2026-09-12.

What an app notice records, and what it refuses to guessA notice holds the model name an app publishes and the date that publication last changed. It does not hold a name inferred from output, a name given in a forum, or a name a support agent mentioned, because none of those can be re-opened by a reader six months later.RecordedDeclinedA published model nameGoes in with its date—Output that looks familiar—A guess, not a statementA forum or a support reply—Not a page a reader can openNo engine named at allRecorded as unnamedNot treated as unchangedTwo columns, one standard of evidence
Fig. 1 Inference would make this register faster to fill and impossible to check, which is the trade it declines to make.

Apps change engines quietly. A product built on one model in March can be running another by September with no announcement, and the only durable evidence is what the app itself published and when. That makes the date as important as the name.

Where an app names no model at all, the notice says so. Running on an unnamed engine is a legitimate product decision and a material fact for anyone who has to plan around a retirement date they cannot see coming.

Notices are rewritten when the seller's own page changes, and the previous reading stays visible rather than being overwritten, because the interesting part of a swap is that it happened.

Three entries here are not applications at all. They are cloud platforms whose lifecycle documentation carries dated retirements and a notice period counted in days or months, which is where every published figure in this register comes from. A fourth publishes the weights themselves, where a withdrawal cannot reach a copy already downloaded.

1The apps

  • SceneMixer — The video models SceneMixer names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • LTX Studio — The video models LTX Studio names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Kling AI — The video models Kling AI names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Pika — The video models Pika names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • invideo AI — The video models invideo AI names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Runway — The video models Runway names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Luma — The video models Luma names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Adobe Firefly — Which video models Adobe Firefly names per plan, whose models they are, which carry an announced end date, and where that date was published.
  • Hailuo AI — The video models Hailuo AI names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • PixVerse — The video models PixVerse names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Vidu — The video models Vidu names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • fal — The video models fal names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Microsoft Foundry — The video models Microsoft Foundry lists with retirement dates, the notice it commits to, and what a deployment does once the date passes.
  • Amazon Bedrock — What Amazon Bedrock publishes about the life of a model it resells: three states, a notice period in months, and what a request does after the date.
  • Gemini API — The video models Gemini API names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.
  • Wan-AI — The video models Wan-AI names on its own pages, whose models they are, which of them carry an announced end date, and where that date was published.

An app is tracked when it publishes a model name somewhere a user can find without an account.

Other notices: Models, Fields, Side by side, Questions, Terms, Learn, Data. Where each line comes from is set out on the sourcing page.