Wan 3.0 is Alibaba’s August 6, 2026 video generation model — and the headline is duration: 30 seconds in a single pass, double the 15-second maximum of Wan 2.7. That’s enough for continuous camera movements, one-take shots, and scene compositions that shorter generators can’t attempt without stitching.
But the more interesting feature is inputs. Wan 3.0 accepts documents (PDFs, PowerPoints, spreadsheets) and web page URLs as creative sources — converting static, text-heavy data directly into dynamic video. No other major video model offers this.
The trade-off: unlike Wan 2.1, 2.2, and 2.7 (which collectively accumulated 6.9 million downloads as open weights), Wan 3.0 is a closed-weight hosted product. Every generated second produces cloud revenue for Alibaba.
Output specs
| Value | |
|---|---|
| Duration | Up to 30 seconds (single pass) |
| Resolution | 480p / 720p / 1080p |
| Audio | Voice and ambient sound generation |
| Languages | Multilingual voice output |
Input modes
Wan 3.0 is an all-in-one model that unifies capabilities previously split across separate Wan models:
| Input type | Formats |
|---|---|
| Text | Text prompts |
| Image | Reference images for character/style consistency |
| Video | Reference clips (1–15s, up to 15s total) |
| Audio | Reference audio (WAV, MP3, 1–15s) |
| Documents | DOCX, DOC, XLS, XLSX, PPTX, PPT, PDF, TXT, KEY, PAGES, NUMBERS, MD |
| Web links | Public web pages (no login required) |
Pricing
| Resolution | Rate per second |
|---|---|
| 480p | $0.05 |
| 720p | $0.10 |
| 1080p | $0.20 |
For a 30-second 1080p clip: $6.00 at list price. Alibaba’s docs also show a limited-time 30% discount on the standard tier and a pricier “prime” tier at $0.068 / $0.14 / $0.28 per second. Compare with FLUX 3 Video’s $0.29/s FHD ($5.80 for 20s) or Veo 3.1’s $0.40/s ($8.00 for 20s).
How it compares
- vs Wan 2.7: Wan 2.7 caps at 15s and requires separate models for different tasks. Wan 3.0 doubles the duration and unifies everything — but at the cost of open weights.
- vs Seedance 2.5: Seedance 2.5 also targets 30 seconds. On Artificial Analysis, Wan 3.0 leads text-to-video (1158 vs 1144) at about a third of the price ($12.00 vs $34.12 per minute), while Seedance 2.5 edges it on image-to-video (1125 vs 1118). Wan 3.0’s unique angle is document/web inputs.
- vs FLUX 3 Video: FLUX 3 maxes at 20 seconds with native audio. Wan 3.0 goes to 30s and accepts document inputs, but FLUX 3 has open-weight plans.
- vs MiniMax H3: H3 offers open weights with native audio. Wan 3.0 has longer clips and document inputs but no open weights.
Limitations
- Still in preview. Alibaba’s API reference labels Wan 3.0 “in preview”; model names, prices and availability differ by region, so check your console.
- No open weights. A strategic departure from the Wan series’ open-weight tradition.
- Thin margins. Its lead over Utopai X, Seedance 2.5 and MiniMax H3 is within the boards’ error bars — see the video Best list for how we weigh that.
For the full video model landscape, see the video models directory.
Best for
- Converting documents and presentations directly into video content
- Long-form single-shot video generation for filmmaking and short dramas
- Marketing and educational video production from existing materials
- Simulation video generation for autonomous driving and robotics training
Pros & cons
Strengths
- 30 seconds in a single pass — double Wan 2.7's 15s, enabling continuous camera movements and one-take shots
- Document and web page inputs — converts PDFs, PPTs, spreadsheets and URLs directly into video (unique capability)
- Ranks #1 on Artificial Analysis text-to-video (Elo 1158) at $12.00 per minute — a third of Seedance 2.5's price
- Competitive pricing at $0.05–$0.20/sec — cheaper than most alternatives at comparable resolution
- Unified model — reference, editing, replication and motion driving folded into one (vs Wan 2.7's separate models)
Limitations
- Alibaba's API reference still labels it "in preview"; availability and model names vary by Alibaba Cloud region
- Closed-weight, hosted product — no self-hosting option (unlike predecessor Wan 2.7's open weights)
- A narrow lead — the top four text-to-video models sit within 17 Elo points, and Wan 3.0 is #5 on image-to-video (1118)
- Two tiers to choose from — standard wan3.0-video and a pricier "prime" version — plus a limited-time discount that will end
Frequently asked questions
When was Wan 3.0 released?
Alibaba opened the public beta on August 6, 2026, with access through Alibaba Cloud Model Studio, Qwen Cloud, and the Wan creation site.
How does Wan 3.0 differ from Wan 2.7?
Three major changes: duration doubled from 15 to 30 seconds in a single pass; document and web page inputs added (PDF, PPT, DOC, XLS, MD, web links); and separate task-specific models unified into one. Wan 2.7 remains available with open weights.
Can I use Wan 3.0's API?
Yes, on Alibaba Cloud Model Studio. Alibaba's documentation (updated October 11, 2026) lists wan3.0-video and wan3.0-video-prime with per-second pricing and a 30-second free quota in Singapore, and models differ by region. Alibaba's API reference still labels the model as in preview, so confirm access in your own console.
How does Wan 3.0 rank?
On Artificial Analysis's blind-vote boards it is #1 on text-to-video (Elo 1158, ahead of Utopai X at 1149 and Seedance 2.5 at 1144) and #5 on image-to-video (1118). The boards use different scales, each anchored at Kling 3.0 Pro = 1000.
Is Wan 3.0 open source?
No. Unlike Wan 2.1, 2.2, and 2.7 (which had open weights with millions of downloads), Wan 3.0 is a closed-weight hosted product. Every generated second routes through Alibaba's cloud infrastructure.
What document formats does it support?
DOCX, DOC, XLS, XLSX, PPTX, PPT, PDF, TXT, KEY, PAGES, NUMBERS, and MD files. Also supports public web page URLs. File size limits apply per format.
Sources
- Artificial Analysis — AA-Video-T2V v2.0 leaderboard
- Artificial Analysis — AA-Video-I2V v1.1 leaderboard
- Alibaba Cloud Model Studio — Wan3.0 video generation API reference
- Alibaba Cloud — Wan3.0 announcement
- RuntimeWire — Wan 3.0 public beta
- TNGlobal — Wan3.0 public beta with 30s clips
- Siray AI — Wan 3.0 explained