Alibaba released Wan3.0, the newest entry in its AI video generation line, on 24 August 2026, and for the first time kept the model’s weights closed. That decision breaks with the open approach that built the Wan series’ reputation among developers, and it landed alongside a $10.2 billion share placement in Hong Kong earmarked for AI investment, according to Dataconomy. The pairing signals where Alibaba’s video ambitions are headed: toward a paid, gated product rather than a freely downloadable research release.
The closed-weight shift matters more than any single feature in this launch. Wan2.7 and its predecessors shipped as open weights, letting outside developers fine-tune and redistribute the models, a strategy that helped Alibaba compete for attention against closed labs. Wan3.0 reverses that. Access now runs through an application process on Alibaba Cloud’s Model Studio and the Qwen Cloud platform, with a members-only consumer site at wan.video planned but not yet open. Dataconomy ties the reversal to a push to monetize Alibaba’s AI capabilities directly, rather than cede downstream value to third-party builders.
Wan3.0 entered public beta on 6 August 2026, ahead of this broader rollout. It generates a single continuous clip of up to 30 seconds, twice the 15-second ceiling of Wan2.7, at resolutions up to 1080p. The company claims the longer clip keeps a subject’s look steady, holds the scene’s geometry in place, and avoids drift in movement from first frame to last, while syncing small expressions on a face and speaking in more than one language. None of those quality claims have been independently benchmarked. They rest on Alibaba’s own description of the model, a gap Dataconomy notes directly.
Inputs extend well past a text prompt. The model accepts images, audio, existing video, web pages and documents such as PDFs and slide decks, which means a marketing team could hand it a finished presentation and receive a narrated video in return. Dataconomy frames that document-to-video path as aimed at corporate communications and marketing teams, not at individual creators experimenting with prompts.
Pricing scales with resolution: $0.05 per rendered second at 480p, $0.10 at 720p and $0.20 at 1080p, which totals $12 for one minute at the top tier. Dataconomy sets that against Google’s Veo 3.1, priced at $0.40 per second for standard service, a rate roughly double Wan3.0’s top-tier cost before any volume discount. That gap is the clearest evidence of Alibaba’s monetization intent: undercut the incumbent on price while restricting who can build on top of the model itself.
The launch arrives against a financial backdrop that explains the urgency. Alibaba’s cloud and AI revenue rose 45 percent year over year to 48.44 billion yuan in the June 2026 quarter, even as quarterly net profit dropped 75 percent, a decline the company attributes to capital expenditure that reached 67.68 billion yuan. Alibaba committed 380 billion yuan to AI infrastructure over three years starting in early 2025, and reports cited by Dataconomy suggest that figure could rise to 480 billion yuan.
Read together, a video model with tighter access and a $10.2 billion capital raise look like two halves of the same bet. Alibaba is spending heavily on AI infrastructure and needs a paid product, not a giveaway, to show a return on it. Teams currently building on open Wan weights should plan a migration path now: nothing in this release suggests Alibaba intends to reopen future versions of the model.
Kerem Gülen reported this for Dataconomy on 24 August 2026.
