Chinese AI companies have captured eight of the top ten positions on a text-to-video ranking that includes audio, according to a new comparison reported by the South China Morning Post. Alibaba’s Wan 3.0 sits first on the Artificial Analysis leaderboard, ahead of Google’s Gemini Omni Flash. Models associated with MiniMax and ByteDance are also near the top, making video generation one of the clearest areas in which several Chinese vendors are competing simultaneously with U.S. counterparts.
The South China Morning Post reported the ranking on August 29. A fal-tuned MiniMax H3 variant took third place. The open-weight H3 followed, with ByteDance’s Seedance 2.0 in the next position. The positions are based on Artificial Analysis’ text-to-video-with-audio leaderboard, which uses blind user comparisons with identical prompts. Rankings can move as models are updated, so they should be read as a snapshot rather than a permanent statement of capability.
The breadth of the Chinese presence is as striking as the first-place model. AI video is no longer a contest between one national champion and one U.S. laboratory. Alibaba, MiniMax and ByteDance have all built products around different combinations of video generation, audio, open weights and commercial distribution. That gives China a deeper bench in an area where training data, product cycles and user demand can matter as much as raw compute.
Wan 3.0 Leads the Text-to-Video With Audio Ranking
SCMP reports that Wan 3.0 was first on the Artificial Analysis text-to-video-with-audio ranking at publication, and Google Gemini Omni Flash took second. The separately listed fal-tuned H3 result shows another feature of the market: a Chinese base model can gain global distribution and further tuning through an overseas platform while the original MiniMax H3 remains available as open weight.
ByteDance’s Seedance 2.0 is among the Chinese models named near the top. Together, the entries from Alibaba, MiniMax and ByteDance occupied eight of the top ten places. SCMP also says Chinese models lead across related image-to-video and video-editing comparisons. The source’s conclusion is based on a specific evaluation platform and its methodology, not a universal measure of every kind of video-generation ability. A benchmark may emphasize some prompts or visual criteria more than others, and product users will have different needs for control, consistency, safety or licensing.
Still, the rankings are commercially relevant because video tools are evaluated by people generating content, not only by static technical tests. Blind comparisons using the same prompt can show whether users prefer a model’s output without knowing which company made it. That format does not eliminate every limitation, but it creates a visible contest in which companies can update models and observe how their results compare.
Alibaba’s first place also gives new context to its August 25 release of Wan 3.0. EastFrontier reported then that Wan3.0 added document-based video input. The latest leaderboard discussion is not another product announcement. It measures Wan 3.0 against a wider field that includes Google, MiniMax and ByteDance.
Short-Video Data and Pricing Shape China’s Video AI Contest
SCMP attributes China’s strength in video generation to several factors: proprietary training data, fast product iteration, large domestic demand for short-form video and aggressive pricing. These are industry explanations reported by the newspaper, not experimentally proven causes of every ranking result. They are nonetheless plausible commercial advantages in a category where companies need both a model and a steady flow of user feedback, content formats and distribution channels.
China’s short-video ecosystem gives companies access to an enormous market of creators, advertisers and viewers accustomed to making and consuming brief visual content. A video model can be tested against practical requests for product clips, social-media posts, avatars and edited footage. That does not solve copyright, quality or safety questions. It does create a setting in which an AI vendor can learn quickly which outputs users find useful.
Pricing is another part of the competition. SCMP said Chinese providers are using lower prices to gain traction. Cost determines whether video generation remains an occasional creative experiment or becomes a regular production tool for a small business. For a developer integrating a video model into an app, a lower per-second cost can affect whether the product can be offered freely, subsidized or sold at a predictable price.
MiniMax’s position in the ranking is particularly notable because its original H3 and the fal post-trained version are treated separately. The company’s recent financial reporting showed that it is pursuing a commercial path across models and enterprise services. EastFrontier’s August 27 story on MiniMax’s enterprise AI revenue described that broader strategy. The video leaderboard adds a consumer and creator-facing dimension to the same company’s competitive position.
The Benchmark Is a Snapshot of an Intensifying Market
The eight-of-ten result does not mean Chinese models will remain ahead in every future ranking. Google, OpenAI, Anthropic and other competitors can release new systems, while Alibaba, MiniMax and ByteDance can update their existing products. The most important part of the result is not a claim of final victory. It is evidence that Chinese vendors now have several products capable of competing at the visible top of a fast-moving video market.
The category also demonstrates why model competition cannot be reduced to language benchmarks. Video systems require visual coherence, motion, editing behavior, audio integration and responsiveness to a prompt. They have to operate within a product experience that creators can use. A company with a strong language model is not automatically strong at video; equally, a video leader needs more than a high leaderboard position to create a profitable and trusted service.
Chinese companies are pursuing this field with different commercial structures. Alibaba has Wan and cloud distribution. ByteDance has a large short-video ecosystem and Seedance. MiniMax offers H3 in open-weight and platform-mediated forms. Kuaishou’s work with Shanghai Jiao Tong University on Arabic voice AI shows another Chinese internet company extending AI beyond text into language and media tools for a particular market. The ranking makes clear that video is now an area in which those broader media capabilities can translate into model competition.
For developers and creators, the practical question is which model best matches a particular workflow, not simply which model tops a list on one date. The latest Artificial Analysis ranking gives Chinese AI vendors a strong data point in that decision. With eight entries in the top ten and Wan 3.0 in first place at publication, China’s video-AI companies have turned a domestic short-video advantage into a visible global challenge. The next test will be whether they can sustain that position as models improve, licensing questions sharpen and users demand more controllable results.