LLMSwaps

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

Where the money goes when the engine changes

The rate difference is the smallest line. A migration costs re-establishing a look, re-measuring how many attempts a shot takes, and sometimes regenerating delivered episodes so a season matches. Those three are the budget, and only the first is obvious in advance. As of 2026-09-12.

Where the cost of a migration actually sits. Recorded 2026-09-12.
LineWhy it is easy to underestimate
Re-establishing the lookPrompt and reference work, priced as generation
Re-measuring the attempt ratioThe old ratio silently invalidates the budget
Regenerating earlier episodesOnly needed if a season would otherwise straddle
The rate difference per secondThe one line everybody starts with
Re-approval downstreamExtra passes on the first deliveries after

Inclusion rule. Cost lines a production meets when the model underneath is replaced. Order. From the largest line to the smallest, by experience rather than by rate.

1The attempt ratio moves in either direction

A new version changes how many attempts a shot needs, and the per-second rate may have moved with it. A budget built on the old ratio is a guess until it has been re-measured on real work.

Ten shots is usually enough to know whether the number moved materially. It is a cheap measurement compared with discovering the answer at the end of a production.

The rate difference is the smallest lineA migration costs re-establishing a look, re-measuring how many attempts a shot takes, and sometimes regenerating delivered episodes. Only the rate difference is obvious in advance, and it is the line that matters least.Where the money actually goesRe-establishing the lookReference and prompt work, billed exactly like production andproducing nothing deliverable.Re-measuring the attemptratioTen shots is enough to know whether the budget still holds.Regenerating earlierepisodesOnly if a season would otherwise straddle the change. Adecision, with numbers.The rate differenceThe line everybody starts with, and the one that moves the totalleast.References first, then prompts: tuning against references about to change means tuning twice
Fig. 1 A planned migration costs a fraction of a forced one, which is the practical argument for caring about notice.

2Re-establishing the look is generation work

Getting a cast and a grade back to where they were means regenerating reference material and testing against it, which is billed exactly like production. It is the line people forget because it produces nothing deliverable.

Doing it in the right order matters: references first, then prompts. Prompts tuned against references that are themselves about to be replaced have to be tuned twice.

3Regeneration is a decision, not a consequence

Where a change lands mid-season, the choice is to accept a visible seam or to regenerate the earlier episodes. The second is expensive and sometimes cheaper than living with the first.

Either way it should be decided with numbers rather than discovered at delivery. That requires knowing which episodes came from which engine, which is the per-file record again.

4A planned migration costs a fraction of a forced one

With warning, the testing happens alongside ordinary work and the season boundary is chosen. Without it, everything happens at once under time pressure, and the first attempt at the new look goes out as a delivery.

That is the entire practical argument for caring which entries publish a notice period. The column records who does.

5Where each vendor's own wording lives

None of the above is a date. Announced end dates, and the apps still carrying the models they apply to, are kept on the calendar with the announcement each one came from.

Guidance rather than a dated entry. No line here is a vendor statement, and none should be read as an announcement. The sourced material is on the calendar. Related: Watching a page, Before committing a season.