Which models fal carries
fal (fal.ai) sells access to other companies' video models one model at a time, with a price against each name and no subscription wrapping them, and states that video is billed by output unit rather than by time in the tool. As of 2026-09-12.
| Model | Whose model | Announced end |
|---|---|---|
| Wan 2.5 | Another vendor | No date announced |
| Kling 2.5 Turbo Pro | Another vendor | No date announced |
| Veo 3 | Another vendor | Earlier Veo versions shut down 30 June 2026 |
| Ovi | Another vendor | No date announced |
Inclusion rule. Models the app names on its own public pages. A model the app does not name is not listed, and an end date is recorded only when the vendor that makes the model has announced one. Order. As the app orders them on its own page.
1How many models each application names
| Entry | Rows it publishes |
|---|---|
| Adobe Firefly | 5 rows |
| Amazon Bedrock | 3 rows |
| fal | 4 rows |
| Gemini API | 5 rows |
| Hailuo AI | 1 row |
| invideo AI | 5 rows |
| Kling AI | 3 rows |
| LTX Studio | 4 rows |
| Luma | 6 rows |
| Microsoft Foundry | 2 rows |
| Pika | 1 row |
| PixVerse | 4 rows |
| Runway | 5 rows |
| SceneMixer | 5 rows |
| Vidu | 2 rows |
| Wan-AI | 5 rows |
Inclusion rule. Entries whose own pages name a model, a version identifier or a lifecycle state. House tier names with no model behind them are not counted. Order. Alphabetical by entry name.
| Entry | Rows with an announced end |
|---|---|
| Adobe Firefly | 1 announced |
| Amazon Bedrock | 3 announced |
| fal | 1 announced |
| Gemini API | 5 announced |
| Hailuo AI | 0 announced |
| invideo AI | 1 announced |
| Kling AI | 0 announced |
| LTX Studio | 0 announced |
| Luma | 1 announced |
| Microsoft Foundry | 2 announced |
| Pika | 0 announced |
| PixVerse | 0 announced |
| Runway | 0 announced |
| SceneMixer | 0 announced |
| Vidu | 0 announced |
| Wan-AI | 0 announced |
Inclusion rule. Counted only where the vendor that makes the model has announced a date in public. A date reported by press coverage alone is not counted. Order. Alphabetical by entry name.
2Per-model pricing makes the register's job easy and the customer's harder
An entry that prices each model individually publishes the cleanest menu in this register: there is no tier to hide behind, and a model that leaves has to leave the price list. That is the best case for anyone trying to match an announcement to an invoice.
It is also the arrangement with the least cushioning. A subscription can absorb a withdrawal by pointing the same plan at a different engine; a price list cannot, so the customer feels the change directly, in code, on the day. Neither arrangement comes with a stated notice period at this entry.
3An access layer holds none of the schedules it resells
Every line here was made by somebody else. One of them has already had sibling versions given dated shutdowns by their maker, which is recorded on the Gemini API entry, and that is the shape of how a date reaches a customer of this kind of service: not from the service, from three companies upstream.
The register keeps those two readings apart on purpose. What this page says it sells is one fact with one date. What the makers say about the life of each line is another, and the calendar is where the second kind is collected. A reader who only watches the access layer will see a menu that looks stable until the moment it is not. One more entry here works the same way at a much larger scale: Amazon Bedrock.
4Read column by column
This register reads fal along stored work and default tier. One page each.
- Stored work — a meter per output unit rather than a library, so delivery is the whole relationship
- Default tier — each call names a model and its own price, so there is no default to have
5Sources
Menu taken from the pricing page on fal.ai, 2026-09-22. Any end date beside a model comes from whoever makes it, and is listed on the calendar.
- How the meter worksVideo models are billed by output unit, per second or per video depending on the modelquoted from the page