LLMSwaps

Announced end dates for video models, and which apps still carry them

Two stages with figures against three states you can query

These are the only two entries in the register with a real lifecycle vocabulary, and they built it differently. One publishes two stages, and the stage decides how many days of notice a retirement owes. The other publishes three states and exposes the current one through its interface. From 2 source pages, read 2026-09-22; no maker pays to be listed.

What each vocabulary names, and what it is wired to. Recorded 2026-09-22.
On this pointMicrosoft FoundryAmazon Bedrock
How many conditionsTwo stages, plus retiredThree states, named on the page
Where the current value livesIn the model's documented stageOn the model, readable by software
What the condition decidesThirty or sixty days of noticeWhether new adoption is allowed
What ends the last oneRequests return 410 GoneRemoval from all regions
Can the condition be checked automaticallyNot describedYes, through the interface

Inclusion rule. Both cells on a row come from the entry's own lifecycle documentation. A state named for a non-video model still counts, because these policies cover every model served. Order. Number of conditions first, then what each one is wired to.

1One vocabulary is wired to days, the other to software

Stages that set a notice period tell a production what it is owed before anything happens. States exposed through an interface tell software when to start worrying. Those are different kinds of usefulness and neither entry has both.

A production would want the first and a platform team would want the second. Most readers of this register are the first and are two layers away from either.

2Three conditions describe a process; two describe a status

Naming a middle state with a period attached turns a removal into something with a beginning. Two stages plus retirement describe where a model sits rather than how it leaves, and the leaving is covered by the notice policy instead.

Both end up with a period and a date. They arrive at it from opposite directions, which is why the register keeps the state column separate from the notice column.

One vocabulary wired to days, the other to softwareStages that set a notice period tell a production what it is owed before anything happens. States exposed through an interface tell software when to start worrying. Neither entry has both.Two stagesThree statesWired toDays of notice owedWhether software should actRead byA person, in a policyCode, through the interfaceDescribesWhere a model sitsHow a model leavesEnds with410 Gone on the dateRemoval from all regionsThey arrive at a period and a date from opposite directions
Fig. 1 Both are platforms selling to developers; twelve applications publish no condition of any kind.

3Neither is an application

These are the two most complete lifecycle vocabularies here and both belong to platforms selling to developers. Twelve applications publish no state of any kind.

That pattern is the finding rather than the details of either vocabulary. The notice question examines why the commitment appears on one side of the market.

4The same question, put to a third entry

What SceneMixer publishes on this column, beside Microsoft Foundry and Amazon Bedrock. Recorded 2026-10-13.
On this pointSceneMixerSource
A tier briefly unavailableA tier that is temporarily unavailable is hidden from the tier list rather than offered and refusedSceneMixer, withdrawn tier guide
A tier that has been retiredA segment submitted on a retired tier is refused with a message before anything is held, and the selector no longer lists that tierSceneMixer, tier no longer exists guide

Inclusion rule. SceneMixer's own statements that bear on the column this pair was chosen for. Order. The statement closest to the column first.

All of its readings are on the SceneMixer entry.

5Sources

Both readings come from the pages the entries publish themselves: Microsoft Foundry and Amazon Bedrock, read 2026-09-22. The column being compared is Lifecycle states. Other pairs: Sora and the Gemini API, LTX Studio and Kling AI.