Alibaba's Wan 3.0 Promises 30-Second Video. Testing Found a 15-Second Wall

The 30-second claim comes with an asterisk

Alibaba put Wan 3.0 into public beta on August 6, 2026, pitching it as the first model to break the 30-second single-pass video barrier, with document-to-video and spreadsheet-to-video generation as its headline trick. On paper, that's a real jump. Most competing video models still cap out well under half a minute per generation, so a genuine 30-second pass would be worth the attention it's getting.

Independent testing from Atlas Cloud found the real number is more complicated. Three undocumented limits cap the longest genuinely continuous clip at 15 seconds, and chaining a second continuation onto that clip makes the model loop backward instead of moving forward in time. The reviewer's best unbroken run at 1080p was 15 seconds: a push in on a neon sign, a vendor closing up a rooftop coffee stall in the rain, then a crane down to a wet street and a taxi. Good footage, just half the advertised length.

What Wan 3.0 actually ships with

The pricing is the part that should get more attention than it has. $0.20 a second at 1080p undercuts Veo 3.1's Standard tier, which runs $0.40 a second, by half. Even accounting for the real-world 15-second ceiling, that's still an aggressive price for the quality on display.

The open-weights question everyone's watching

Alibaba's Wan line built its reputation on open releases: Wan 2.2 shipped with open weights, and a lot of the model's early goodwill came from developers who could actually run it themselves. Wan 3.0 breaks that pattern. It's cloud-only, gated behind an application, with no committed date for open weights and a company history here that cuts both ways: Alibaba has open-sourced before, but it's also walked back open promises before. The community reaction described by outlets covering the launch is cautious enthusiasm mixed with a wait-and-see attitude specifically on this point, not the immediate excitement a fully open release would have gotten.

Where it actually fits

Nobody currently wins on every dimension in AI video. Wan 3.0's own comparisons put it well ahead of Runway Gen-4.5 and Grok Imagine Video on the company's internal testing, and it's the clear price leader against Veo 3.1. But the document-to-video and spreadsheet-to-video modes are the more interesting story than the duration claim, since nothing else in the current field treats a spreadsheet as raw input for a video generation model. If your use case is turning structured data into an explainer clip rather than a pure creative render, that's a real departure from what most of the competition is building toward. If video specifically isn't the point and you just need strong still images or vector art, tools like [Midjourney](https://questloops.com/tools/midjourney) or [Recraft](https://questloops.com/tools/recraft) are still the more direct route.

Should you try it

If you can get access, the pricing alone makes it worth testing against whatever you're using now, especially for shorter clips where the 15-second reality doesn't cost you anything the marketing promised. Just don't plan a project around a clean 30-second single take until Alibaba or an independent source confirms that duration holds up outside a demo reel. And don't expect to self-host it: this is a cloud-only product for now, unlike the Wan releases that built the brand's reputation in the first place.