Skip to main content
Tools5 min readNovember 6, 2026

Open vs Closed AI Video Models: What Is Actually Open

Wan 3.0 is closed. LTX-2.5 is gated with a revenue limit. Which AI video models you can run yourself as of November 2026, and the licence questions to ask.

S

StudioList Editorial

AI Video Research Team

Most of the AI video models people call open source are not. As of November 6, 2026, Alibaba's newest model, Wan 3.0, is closed and API-only, and the best-known current open release, LTX-2.5, is gated and has a revenue limit in its licence. Older models with permissive licences still exist, but they are older.

Three different things

  • Open source usually means a licence such as Apache 2.0 that lets you use, change and sell with few conditions.
  • Open weights means you can download the model. The licence can still restrict who may use it and how.
  • Closed or API-only means you pay per second and the model runs on the vendor's servers.

"Free" on a model page often means only that the weights cost nothing to download. You still pay for hardware and the time to run it.

Where each model stands

ModelStatusLicenceSource
Wan 3.0 (Aug 2026)Closed, API-onlyCommercial APIDataNorth
Wan 2.2 (2025)Open weights, last open Wan as reportedApache 2.0DataNorth
LTX-2.5 (reported Aug 2026)Open weights, gated on Hugging FaceLTX-2.x Community License: free commercial use under $10M annual revenue, paid above. Not OSI open sourceHugging Face
HunyuanVideo-1.5 (Nov 2025)Open weightsWe did not confirm the terms. Read the repo licenceGitHub
Mochi 1 preview (Oct 2024)Open weights, 480p. We found no newer releaseApache 2.0Hugging Face

Two entries we would not rely on yet. We could not confirm the status or licence of Wan 2.7, which earlier coverage, including our own April guide, described as open source. We also could not confirm the maker or licence of HappyHorse 1.0. It is listed on Runway's API, and its open-weight claims are disputed.

Why this matters to a buyer

Open weights can be a real advantage. A studio can run the model on its own machines, train it on your product or character, and keep your unreleased material off a vendor's servers. It also gets no per-second bill.

It does not settle everything. A studio running local models still needs to answer for what the model was trained on, the licence terms, and whether the output is good enough for your deliverable.

Questions to ask a studio

  1. Which exact model and version will you use, and is it running locally or through an API?
  2. What is the licence, and do its revenue or use limits apply to you, to us, or to neither? Get the answer in writing.
  3. If our footage or product images go into the pipeline, where do they go and who can keep them?
  4. If you train a custom model on our material, who owns it and what happens to it after the project?

We are a directory and this is not legal advice. For a broadcast campaign or anything with real legal exposure, have counsel read the licence.

For closed models, specs and prices, see AI video models in November 2026.

Ready to find the right studio for your project?