Update — August 28, 2026: In fal's published latency comparison, Wan 3.0 averages 176.93s against H3 Max's 3.49s, at 2044 vs 2080 Bayesian Elo. Wan 3.0 still wins on duration (30s) and resolution (1080p).
August 24, 2026 — Alibaba Cloud moved Wan 3.0 from beta to general availability: a closed-weight video model that generates up to 30 seconds in one pass — and, unusually, accepts office documents and URLs as prompts, not just text and pixels.
If you build marketing automations or demo videos, the question is whether $0.20/second at 1080p beats your current screen capture + editor loop — or whether you should route through agentic video stacks instead.
TL;DR — what people are asking
| Question | Answer |
|---|---|
| GA date? | August 24, 2026 (beta since Aug 6) |
| Max length? | 30 seconds single generation + extension mode |
| Inputs? | Text, image, audio, video, docs, webpages |
| Resolutions? | 480p / 720p / 1080p |
| API ID? | wan3.0-video on Model Studio / Qwen Cloud |
| Open weights? | No — API only |
| 1080p × 30s cost? | ~$6 at list $0.20/s (check promos) |
Signature capability: document-to-video
Alibaba's cloud blog emphasizes Omni-Reference inputs — maintain faces, logos, and style across scenes when you supply reference media.
The builder-facing twist is direct document ingest:
- Product planning slides → promotional cut
- PDF spec → narrated walkthrough
- Webpage URL → social clip without manual scraping
Limits: one file or link, ≤100MB, ≤50 pages per request.
That targets teams who already have collateral in Google Slides or Notion exports — not filmmakers storyboarding from scratch.
Pricing math for practitioners
| Tier | $/second | 15s clip | 30s clip |
|---|---|---|---|
| 480p | $0.05 | $0.75 | $1.50 |
| 720p | $0.10 | $1.50 | $3.00 |
| 1080p | $0.20 | $3.00 | $6.00 |
Alibaba ran a 30% Standard-tier discount through September 23, 2026 on some announcements — confirm in Model Studio billing before quoting clients.
Compare to Gemini video workflows and local GPU time on Mac Studio if you batch hundreds of variants.
What this means for what you build or pay
Marketing ops: Wan 3.0 is a API SKU, not a repo — budget like any cloud inference line item. Good for deck→video pipelines wired into Qwen Cloud credentials you may already hold for Qwen 3.8.
Product demos: One-shot 30s limits mean you still chunk long tutorials — pair with Wan 2.7 carryover edit-in-place features (modify dialogue/plot without full regen per Alibaba docs).
Agent builders: For reproducible multi-scene work, keep ViMax-style orchestration; use Wan as a tool node inside a harness when you need Alibaba's document reader.
Wan 3.0 vs nearby video models (Aug 2026)
| Model | Max single pass | Doc input | Weights | Typical use |
|---|---|---|---|---|
| Wan 3.0 | 30s | Yes | Closed API | Decks → ads |
| Seedance 2.5 | 30s | Limited | Closed API | Multi-ref clips |
| ViMax agents | Variable | Via tools | Harness-dependent | Storyboard pipelines |
Independent benchmarks were thin at GA — treat comparisons as demo quality, not leaderboard gospel.
Honest limitations
- Closed weights — no air-gapped deploy; China/US data residency rules apply.
- 30s cap — long-form still needs editing or extensions.
- No 4K tier published at GA.
- Reuters tied launch to share sale — ignore ticker noise; evaluate API on your assets only.
- On-screen text accuracy — Alibaba says they're still improving typography — verify legal disclaimers manually.
Related on explainx.ai
- ViMax agentic video generation guide
- Qwen 3.8 Flash-Next release
- Can LLMs watch video?
- How to generate videos with Google Vids/Veo
- Palmier Pro AI video editor MCP
- Create product demo videos with Claude
- China AI playbook — open weights
- Blur faces in video — AI privacy
Update — October 6, 2026: Open-weight rival with synced audio: Kandinsky 6.0 Video (29B, MIT).
Wan 3.0 API pricing and document limits per Alibaba Cloud announcements as of August 26, 2026 — confirm current rates in Model Studio before production spend.