The one successor named anywhere in this register
The retirement schedule carries two dated rows for the hosted video model. The earlier one names its replacement, a later version of the same model. The second names none. This is the only place in the register where a retirement and a destination were published in the same sentence. As of 2026-09-22.
| What the page is asked | What Microsoft Foundry publishes | What that leaves open |
|---|---|---|
| Does a retirement name what follows | Yes, for the earlier of two rows | Within the same model family, not across makers |
| What the successor actually is | A later version of the same model | A version bump rather than a different engine |
| Does the second row name one | No replacement is named for it | And its date is later than the first |
| Is the move performed for you | Not stated on the schedule page | A deployment names a version, so somebody must |
Inclusion rule. Read from the retirement schedule table, restricted to the video rows. A successor mentioned in a launch announcement but not in the schedule does not fill a cell. Order. What the schedule states first, then what it leaves to the customer.
1A successor inside the same family is the easy case
Naming a later version of the same model is the least demanding form of this disclosure. Nobody had to choose between makers, and the behaviour a caller depends on is likely to be closer than it would be under any other move.
It is still the only one published here, which says more about the field than about this platform. Everywhere else the ending arrives without a destination, and a reader has to infer the obvious candidate and hope.
2The second row shows the pattern breaking immediately
The successor named for the first row is itself scheduled to retire, and its own row names nothing. So the chain runs one link and stops, which is a more honest picture of what a version successor buys than a single row would give.
A production reading only the first row would conclude that this platform manages succession. Reading both shows that it documents one step at a time, and that the step after the next one has not been written yet.
3Naming is not moving
Nothing on the schedule says the platform performs the change. A deployment names a version explicitly, which is what makes the retirement visible in the first place and also what guarantees somebody has to edit something.
That puts this entry in the same position as Amazon Bedrock, which states the point outright. The difference is that this one at least says where to go.
- sora-2 version 2025-10-0615 July 2026retirement date, replaced by sora-2 version 2025-12-08
- sora-2 version 2025-12-0815 October 2026retirement date, no replacement named
- Whether a date can be pushed back on requestRetirement dates are not extendable
4Sources
Read from the model retirement schedule on learn.microsoft.com, 2026-09-22. The whole entry is on Microsoft Foundry, and the column it sits in is on Named replacement. Same column, other entries: Sora, Gemini API.