Tell a complete story in one thirty-second pass with Wan3.0.
Wan3.0 is Alibaba’s current video model. A single generation can run to 30 seconds — where the model it replaces stopped at 15 — and its reference inputs include documents and webpages, not only text and pictures.
Capabilities and specifications on this page come from Alibaba’s Wan product pages, its Model Studio documentation and the Wan3.0 general-availability announcement. Prices are set by the vendor and change without notice.
What Wan3.0 renders
These frames are the stills Alibaba publishes for Wan3.0, taken from the posters of its own demo videos. Between them they cover the ground the release is sold on: simulated light, action choreography, and interiors with real depth.
Example images belong to Alibaba and are reproduced with credit: Wan official site
How Alibaba describes Wan3.0
Alibaba sells Wan3.0 on one number — thirty seconds in a single pass — and on a new class of input. The vendor’s own announcement is unusually specific about both, including what has not been finished yet.
Thirty seconds is the differentiator
Wan2.7 ran 2 to 15 seconds; Wan3.0 generates up to 30 in one pass. That buys room for narrative pacing and continuous camera movement rather than a single shot, and two features make the longer canvas practical: the model recommends a duration from the prompt, and a clip can be extended.
Everything to video
Beyond text, images, audio and video, Wan3.0 reads documents and webpages — DOC, XLS, PPT, PDF, TXT, KEY, Pages, Numbers and Markdown, one file or link per request, up to 100 MB and 50 pages. A product deck becomes a launch film; a spreadsheet becomes animated charts.
Consistency as a product requirement
Alibaba frames the upgrade around identity, voice and lip sync holding across a sequence, so characters, props, spaces, lighting and style stay aligned through an action sequence or a multi-language performance.
The vendor’s own caveat
Alibaba’s general-availability write-up says plainly that audio texture and on-screen text rendering accuracy are still maturing. That is a narrower claim than the marketing pages make, and it is the one to read before committing a deliverable.
Per-second pricing by resolution
Alibaba prices Wan3.0 by the second of generated video, at rates that depend only on resolution. These are the standard list prices on the international channel since general availability on 24 August 2026.
| Resolution | Billing | List price |
|---|---|---|
| 480P | Per generated second | $0.05 |
| 720P | Per generated second | $0.10 |
| 1080P | Per generated second | $0.20 |
A 30-second draft at 480P works out at about $1.50 and a 30-second finish at 1080P at about $6.00, before any promotional discount. Alibaba’s China platform lists the same three tiers at ¥0.30, ¥0.60 and ¥1.20 per second.
What Wan3.0 is built to do
These are the capabilities Alibaba lists on the product page, with the vendor’s own wording.
Reference to video
Cast referenced characters into new scenes, supporting multi-person and human-object interaction with consistent appearance and voice.
Image to video
Alibaba’s wording: generate videos up to 30 seconds with automatic scene splitting, for steadier native audio and picture.
Text to video
Cinematic precision with sophisticated motion and highly reliable prompt execution.
Video editing
Multi-dimensional, instruction-based editing that reworks visuals, storylines and environments without regenerating the whole concept. Wan2.7 introduced it and Wan3.0 carries it forward.
Cinematic control
Lighting, colour grading and composition are tuned to cinematic standards, with complex action choreography and physics simulation called out by name.
Immersive audiovisual
Binaural audio with support for regional dialects, cut to the rhythm of the picture so sound and image breathe together.
Documented specifications
Every row below is stated in Alibaba’s Model Studio documentation or its general-availability announcement.
- Developer
- Alibaba
- Current model
- Wan3.0
- Model ID
- wan3.0-video
- Single generation
- Up to 30 seconds
- Output resolutions
- 480P, 720P, 1080P
- Duration help
- Smart duration recommendation and video extension
- Reference inputs
- Up to 10 images, 5 video clips and 5 audio tracks
- Document inputs
- DOC, XLS, PPT, PDF, TXT, KEY, Pages, Numbers, Markdown
- Document limits
- One file or link per request, up to 100 MB and 50 pages
- Audio
- Generated with the video, including lip sync
- Video editing
- Built in; carried forward from Wan2.7
- Weights
- Not published
Documented access channels
Wan3.0 reached general availability on Alibaba Cloud’s international channels on 24 August 2026 and is billed by the second. Alibaba also publishes downloadable weights for the earlier models in the family.
Wan official site
Alibaba’s product page for the model, including the capability list quoted on this page.
OpenWan3.0 on Model Studio
The model’s documentation on Alibaba Cloud’s platform: input modalities, the three resolutions and the 30-second ceiling.
OpenGeneral-availability announcement
Alibaba’s own write-up of the release, with the pricing table and the caveat about audio and on-screen text.
OpenOpen-source releases
Alibaba publishes downloadable weights for the earlier Wan models, up to Wan 2.2 under Apache 2.0. Wan3.0 itself has no published weights, so this repository is the family’s open half, not this model.
OpenThird-party platforms
Wan3.0 is also served through third-party API platforms. The model is the same; the pricing, rate limits and terms are theirs.
Wan3.0 questions
How long can a single Wan3.0 generation be?
Up to 30 seconds in one pass, at up to 1080P. Alibaba says the model analyses the prompt and recommends a duration, and that a generation can be extended to continue a storyline.
Does Wan3.0 make its own audio?
Yes. Native audio and picture are produced together, including lip sync. Alibaba’s own general-availability post notes that audio texture is still maturing, so treat the audio as something to review rather than something to ship unchecked.
Can I feed Wan3.0 a document?
Yes, and that is the genuinely new input class. DOC, XLS, PPT, PDF, TXT, KEY, Pages, Numbers and Markdown are accepted alongside webpages — one file or link per request, up to 100 MB and 50 pages.
How many references can I combine?
Alibaba’s own listing says up to 10 images, 5 video clips and 5 audio tracks in one generation, alongside the document or webpage input.
Are Wan3.0’s weights open?
No. Alibaba has not published weights for Wan3.0; it runs through Alibaba Cloud Model Studio and third-party platforms. The downloadable releases in this family stop at Wan 2.2, which shipped under Apache 2.0.
What does the Wan3.0 API cost?
Standard list pricing on the international channel is $0.05 per second at 480P, $0.10 at 720P and $0.20 at 1080P, so a 30-second 1080P generation is about $6.00 before any discount. Alibaba’s China platform lists the same tiers at ¥0.30, ¥0.60 and ¥1.20 per second.
Can Wan3.0 edit a video I already have?
Yes. Wan2.7 introduced video editing and Wan3.0 carries it forward: visuals, plot and dialogue can be modified without regenerating the concept from scratch.