It wasn't on any leaderboard a month ago. Alibaba shipped Wan 3.0 on August 24, 2026, and it went straight to the top of Artificial Analysis's text-to-video rankings — tied with Google's Gemini Omni Flash. Here's what it actually is, what it costs, and whether it's worth switching to.
Alibaba launched Wan 3.0 on August 24, 2026 through Alibaba Cloud's Model Studio and API, following a public beta that opened August 6. It can generate up to 30 seconds of video in a single pass — roughly double the length its predecessor, Wan2.7, could manage — at up to 1080p, and it accepts an unusually wide range of inputs: text, images, audio, video, documents, spreadsheets, slides, and web pages. The launch came one day after Alibaba raised HK$80 billion (about $10.2 billion) in Hong Kong's largest follow-on share offering, with the company saying proceeds would go toward AI infrastructure — the timing wasn't lost on anyone covering it.
On Artificial Analysis's Video Arena — the same head-to-head, Elo-scored human-preference leaderboard we track on our sourced rankings page — Wan 3.0 debuted at or effectively tied for #1 on the text-to-video board almost immediately, ahead of established models from Google, ByteDance, and MiniMax that have been iterating on that leaderboard for months. That's a genuinely fast climb: for context, Runway's Gen-4.5 led the same board at its own launch in late 2025 and had fallen out of the top 10 within six months as competitors caught up. Whether Wan 3.0 holds the top spot as long remains to be seen — these boards move fast, and we'll update this page and our rankings page when they do.
Two things stand out beyond the raw ranking. First, document-to-video input: feeding it a slide deck, spreadsheet, or web page and getting a video out is not something the other major tools on this site (Runway, Pika, Kling, Seedance, Veo) advertise as a core input type — they're built around text prompts and images. Alibaba is positioning this specifically for marketing, product-demo, and explainer-style use cases where the source material already exists as a document. Second, 30-second single-pass generation — every other major tool caps a single generation at somewhere between 5 and 10 seconds, requiring multiple generations stitched together for anything longer. Wan 3.0 doing that natively, in one pass, is a real capability gap versus the field, not just a marketing number.
| Resolution | Price per output second | ~5-second clip |
|---|---|---|
| 480p | $0.05/sec | ~$0.25 |
| 720p | $0.10/sec | ~$0.50 |
| 1080p | $0.20/sec | ~$1.00 |
List rates via Alibaba Cloud, per-second metered pricing (no subscription tier). At these rates, Wan 3.0 undercuts Seedance's official per-clip pricing (~$0.76 at 720p, ~$1.87 at 1080p — see our full pricing comparison) at every resolution, though a 30-second Wan 3.0 clip at 1080p (~$6) costs more in absolute terms than a short 5-second clip anywhere else — it's a longer video, not a directly equivalent unit.
This is the catch: there's no single polished consumer web app yet, the way Runway, Pika, Kling, and Veo each have one. Wan 3.0 is reached through Alibaba Cloud's Model Studio console or its API — closer in shape to how Seedance is typically accessed than to a type-a-prompt-and-click tool. If you're comfortable with an API key and a metered-billing dashboard, it's straightforward; if you were hoping for a Runway-style consumer app, it's not there yet. Expect third-party playgrounds and resellers (the same kind that already host Seedance — fal.ai, Replicate, and similar) to add Wan 3.0 access over the coming weeks, the way they typically do once a model proves popular.
Once you've generated a clip
Same as every other tool on this site: Wan 3.0 exports a full rendered scene, background included, with no transparent-export option. If you want to lift the subject out to composite it elsewhere or turn it into a sticker or overlay, that's a background-removal step done after generation — RemoveGifBG processes the exported MP4 frame by frame. See the AI video background remover guide.
Related: More AI video news · Full pricing comparison · Why Sora isn't on the leaderboard · Google's Veo pricing restructure
Wan 3.0 is Alibaba's newest AI video-generation model, launched August 24, 2026 through Alibaba Cloud's Model Studio and API, following a public beta that opened August 6. It generates up to 30 seconds of video in a single pass — about double the length of its predecessor, Wan2.7 — at up to 1080p, and accepts text, images, audio, video, documents, spreadsheets, slides, and web pages as input.
Wan 3.0 is priced per output second, not as a monthly subscription: roughly $0.05/second at 480p, $0.10/second at 720p, and $0.20/second at 1080p, per Alibaba Cloud's list rates. A 5-second clip works out to about $0.25 (480p), $0.50 (720p), or $1.00 (1080p) — noticeably cheaper than Runway's or Seedance's per-clip cost at comparable resolution.
On Artificial Analysis's Video Arena — human preference voting converted to an Elo score, not a marketing claim — Wan 3.0 debuted at or near the top of the text-to-video leaderboard, running roughly tied with Google's Gemini Omni Flash and ahead of Veo 3.1, Kling, and Seedance on that specific board. "Better" still depends on what you need: Veo remains the strongest for native audio, Kling for physically plausible motion, and Wan 3.0's own standout feature is document/slide-to-video input and 30-second single-pass length, not any one of those specialties.
Through Alibaba Cloud's Model Studio console or its API — there isn't yet a single polished consumer web app the way Runway, Pika, Kling, or Veo each have one. That makes it, for now, a closer fit for developers and technical users than for someone who just wants to type a prompt and click generate, similar to how Seedance is typically reached.