Wan 3.0 makes up to 30 seconds of video in one pass
Wan 3.0 is Alibaba's video model, and one generation returns up to 30 seconds of footage. Alibaba serves it through Alibaba Cloud Model Studio, the cloud service where developers call its AI models. The announcement post is dated 6 August 2026.
The model takes text, images, audio and video as input. Documents are the new addition.
Alibaba pairs the longer length with two features. Smart duration recommends a length based on the prompt. Video extension lets a user continue a story after the first generation ends.
Alibaba also states where Wan 3.0 sits in the Wan family. The family has shipped eight versions, from Wan1.0 to Wan3.0. Alibaba describes that path as a move from generating assets to full creation, and from content generation to commercial delivery.
Wan 3.0 at a glance, as stated by Alibaba
| Item | Detail |
|---|---|
| Longest single generation | 30 seconds |
| Length features | Smart duration, video extension |
| Input types | Text, image, audio, video, documents |
| Document limit | 1 file or link per request, 100MB, 50 pages |
| Output resolutions | 480P, 720P, 1080P |
| Where it runs | Alibaba Cloud Model Studio |
"It generates up to 30 seconds of video in a single pass, accepts text, images, audio, video, and — for the first time — documents (PPT, PDF, DOC, XLS, and more) as input, and delivers a major leap in real-world fidelity: lifelike, diverse human faces and precise reference-to-video consistency."／"The Wan model family has shipped 8 iterations from Wan1.0 to Wan3.0, evolving from asset generation to full creation, and from content generation to commercial delivery."（intro and From Wan1.0 to Wan3.0 sections）View official source →
Document input takes 9 formats, up to 50 pages per request
Wan 3.0 reads nine document formats: doc, xls, ppt, pdf, txt, key, pages, numbers and md. Each request takes one file or one link, up to 100MB and 50 pages.
Most of these are file types from familiar apps. The doc, xls and ppt formats are Word, Excel and PowerPoint files. The key, pages and numbers formats come from Apple's Keynote, Pages and Numbers, and md is Markdown.
Alibaba says Wan 3.0 is the first model in the family to read documents, spreadsheets and web pages as structured input. Documents are the one input type Alibaba marks as new in this release. When a source file is a PowerPoint deck and the text is needed as plain Markdown, the tool below converts it.
"Wan3.0 is the first model in the family to read documents, spreadsheets, and web pages as structured input."／"Supported input formats include doc, xls, ppt, pdf, txt, key, pages, numbers, and md, with a limit of one file or link per request, up to 100MB and 50 pages."（Everything to Video section）View official source →
Wan 3.0 API pricing: $0.05 to $0.20 per second
The Wan 3.0 API bills by the second of generated video, and the rate rises with resolution. The number before P is the picture's height in pixels, so 1080P is sharper than 480P. Alibaba lists three rates.
Wan 3.0 API rate per second and the cost of a full 30 seconds
| Resolution | Per second | 30 seconds |
|---|---|---|
| 480P | $0.05 | $1.50 |
| 720P | $0.10 | $3.00 (calculated) |
| 1080P | $0.20 | $6.00 |
Alibaba states the 30-second totals for 480P and 1080P. The 720P total is calculated here from the listed rate: 30 seconds at $0.10. The 1080P rate is four times the 480P rate, which is why a 30-second render costs $6.00 against $1.50.
Alibaba presents that gap as a way of working: iterate on drafts at 480P, then finish at 1080P. A related model is covered in FLUX 3 Video: 20-Second Clips With Native Audio.
"At these rates, a full 30-second 1080P generation costs $6.00, and a 30-second 480P draft costs $1.50 — practical for iterating on drafts at 480P and finishing at 1080P."（Wan3.0 API Pricing section）View official source →
Alibaba flags audio texture and on-screen text as weak
Alibaba names two weak areas in its own post. Internal testing found that audio texture and the accuracy of on-screen text rendering are improving but not yet where the company wants them. On-screen text means words drawn inside the video itself.
The gains Alibaba claims lie elsewhere. It describes a leap in real-world fidelity, with lifelike, diverse human faces and precise reference-to-video consistency. Reference-to-video means generating a clip from supplied reference material.
Both the weak spots and the gains are Alibaba's own statements about its model. The weak spots come from its internal testing.
"During internal testing, we identified areas where Wan3.0 still has room to grow: audio texture and on-screen text rendering accuracy are improving but not yet where we want them."／"It generates up to 30 seconds of video in a single pass, accepts text, images, audio, video, and — for the first time — documents (PPT, PDF, DOC, XLS, and more) as input, and delivers a major leap in real-world fidelity: lifelike, diverse human faces and precise reference-to-video consistency."（A Note on Current Limitations and intro sections）View official source →
When was Wan 3.0 released, and is it open source?
Alibaba's Wan 3.0 announcement is dated 6 August 2026, and that is the only date the post gives. The post describes API access through Model Studio and says nothing about downloadable weights. Whether weights will be released is not answered there.
Weights are the trained model files. With open weights, developers can download and run a model themselves. With API access, they call the model on the provider's servers, which is the route this post describes.
The post carries an announcement date and names no separate general-availability date. These are gaps in one announcement, not statements by Alibaba about what comes next. In short, Wan 3.0 as announced makes up to 30 seconds of video per generation, takes nine document formats as input, and costs $0.05 to $0.20 per second through the API.
FAQ
- How long can a Wan 3.0 video be?
- Thirty seconds from a single generation. A smart duration feature reads the prompt and suggests a length, and video extension continues a clip that has already been produced.
- What does the Wan 3.0 API cost?
- Billing is per second of generated video, and the rate depends on the output resolution. A 30-second clip at 1080P comes to $6.00, and the same length at 480P comes to $1.50.
- Can Wan 3.0 turn a PowerPoint deck into a video?
- Yes. Alongside text, images, audio and video, it accepts doc, xls, ppt, pdf, txt, key, pages, numbers and md files. Each request takes one file or link, capped at 100MB and 50 pages.
Related articles
Qwen 3.8 Flash Next: 125B Qwen4 Preview, 6B ActiveQwen 3.8 Flash Next activates 6B of 125B parameters per token, takes 262,144 tokens of context and previews the Qwen4 architecture with sparse attention.
Apple Intelligence launches in China with QwenApple Intelligence comes to iPhones in China with iOS 27.3 in December. Alibaba's Qwen handles some Siri and Writing Tools requests, but Siri AI is left out.
Researchers talk to a Tencent Cloud AI agent fleetAn AI agent fleet on Tencent Cloud made 2,600+ attempts on Amap. Researchers traced it and talked to it via a hidden note. Tencent's side cites a model test.
Claude Fable 5 Back as FT Says Top Model StrugglesThe FT says Anthropic's best model struggles to attract users. Claude Fable 5 was suspended June 12, returned July 1 and moved to usage credits July 7.