Wan
Alibaba CloudWan is Alibaba's family of video generation models. Its current release, Wan 3.0, generates clips of up to 30 seconds with sound from text and up to 20 reference assets and supports instruction-based editing, and is served only through Alibaba Cloud Model Studio. The earlier Wan 2.1 and 2.2 releases, from a 1.3B model to a 27B mixture-of-experts at up to 720p, remain downloadable.
The current release, Wan 3.0, is hosted-only, while Wan 2.2 and 2.1 remain downloadable under Apache 2.0. It scores 1/closed on the current release, and the older open line trails it. A separate availability attribute that would record that gap is tracked in #727.
Openness
1 medium confidence- weights
- closed(Wan 3.0 is served only through Alibaba Cloud Model Studio
- data
- closed(nothing is published for Wan 3.0)
- code
- closed(no code is published for Wan 3.0)
- license
- Proprietary(Wan 3.0 is available only under Model Studio API terms)
- prior-release
- distributed(Wan 2.2 (and 2.1): all checkpoints downloadable under Apache-2.0, with inference code only and training data described but not released)
Wan 3.0, the current release, can only be used through Alibaba Cloud's API, and no weights have been published for it. The previous generation, Wan 2.2, remains downloadable under Apache 2.0. The open line trails the vendor's closed release. A separate availability attribute that would record that gap is tracked in #727.
- https://help.aliyun.com/en/model-studio/wan3-video-generation-api-reference recorded 2026-09-26
Model Studio API reference: "wan3.0-video-prime" and "wan3.0-video" as API model names, "Currently in preview".
- https://huggingface.co/api/models?author=Wan-AI&sort=downloads&direction=-1&limit=100 recorded 2026-09-26
Hugging Face listing of the Wan-AI organization: the Wan 2.1 and 2.2 checkpoints, all public and apache-2.0; no Wan 2.5, 2.6, 2.7 or 3.0 repository.
- https://raw.githubusercontent.com/AlibabaCloud-Official/Wan3.0/main/README.md recorded 2026-09-26
README: "Wan 3.0 is a state-of-the-art AI video generation model powered by Alibaba Cloud", linking the hosted wan.video site and Model Studio; no download.
- https://raw.githubusercontent.com/Wan-Video/Wan2.2/main/LICENSE.txt recorded 2026-09-26
Wan 2.2 LICENSE.txt body: "Apache License / Version 2.0, January 2004", covering the prior, still-distributed release.
- https://ungh.cc/repos/AlibabaCloud-Official/Wan3.0/files/main recorded 2026-09-26
File list of AlibabaCloud-Official/Wan3.0: LICENSE and README.md only; no weights or code.
- https://www.alibabacloud.com/blog/wan3-0-30-second-ai-video-generation-from-any-input_603452 recorded 2026-09-26
"Wan3.0 is the new generation of Alibaba's Wan video generation model family, now available on Alibaba Cloud Model Studio", with per-second API pricing.
Adoption
4 high confidenceAdoption is measured as Hugging Face downloads across the Wan 2.1 and 2.2 checkpoints, most of them for the Diffusers-format builds.
- https://huggingface.co/api/models?author=Wan-AI&sort=downloads&direction=-1&limit=100 recorded 2026-09-26
Organization listing sorted by downloads: Wan2.1-T2V-1.3B-Diffusers 264951, Wan2.2-TI2V-5B-Diffusers 159486, Wan2.2-I2V-A14B-Diffusers 143218 in the trailing 30 days.
Capability
5 medium confidenceWan 3.0 leads the video arena outright, ahead of Gemini Omni Flash and MiniMax H3, but it is a hosted model. The downloadable Wan 2.2 places in the lower half of the same board.
- https://artificialanalysis.ai/video/leaderboard/text-to-video recorded 2026-09-26
Arena FAQ: "Wan 3.0 currently leads the Artificial Analysis Text to Video Arena (without audio) with an Elo score of 1335"; leaderboard entry "Wan 2.2 A14B" (rank 53, Elo 1116.1).
Verified 2026-09-26