Chinese AI Video Models Top Global Rankings as Costs Tumble

Chinese Models Dominate the AI Video Leaderboard

Chinese artificial intelligence companies are pulling ahead of their US rivals in video generation, occupying eight of the top 10 positions on the industry benchmark Artificial Analysis leaderboard as of this week. Alibaba's Wan 3.0 ranks first in the text-to-video category with audio, ahead of Google's Gemini Omni Flash, while versions of MiniMax's open-weight H3 and ByteDance's Seedance 2.0 round out the top five, according to data published Friday by the South China Morning Post.

The rankings mark a striking contrast with the frontier large language model race, where US firms such as OpenAI and Anthropic still lead. Analysts say the video arena rewards a different mix of strengths, including proprietary training data, rapid iteration cycles and the vast volumes of short-form content produced by Chinese platforms, rather than compute alone.

Why China's Video AI Is Pulling Ahead

Industry observers point to the short-video ecosystem as the decisive advantage. ByteDance, the parent of TikTok and Douyin, has turned AI-generated clips into a genuine business line by feeding models with the preferences of more than a billion monthly active users, said Kyle Chan, a fellow at the Brookings Institution's John L Thornton China Centre.

"ByteDance has created a positive flywheel, turning AI-generated short form video into a real industry," Chan said. Early hits signalled strong viewer demand, driving the company to double down on the technology and transform it into a lucrative business.

Domestic demand has reinforced the loop. "China's short-drama and AI-generated comic-drama industries have developed rapidly, with Chinese production teams also supplying content for overseas markets," said Wang Zebin, an investment manager at private equity firm JG Investment.

The Copyright Divide Between Beijing, Washington and Tokyo

Copyright policy has created a distinct playing field, analysts said. US AI developers face "much more potential risk and liability" over intellectual property, according to Chan, while comparatively flexible copyright enforcement in China gives domestic firms more runway to test and commercialise their models, Wang said.

The practical difference surfaced in a widely shared example from Shenzhen-based AI enthusiast Cecilia Huang, who said Google's Gemini rejected her request to render a photo into a Ratatouille-style character for a 15-second video on copyright grounds, while ByteDance's Doubao fulfilled the request. ByteDance did not respond to a request for comment on Tuesday, and told the South China Morning Post in February that it "respects intellectual property rights" and was "taking steps to strengthen current safeguards" against unauthorised use of copyrighted content in training Seedance 2.0.

For Japan, the comparison is instructive. Tokyo has charted a middle course: Japanese guidelines published in the first half of 2026 by the Japan Patent Office and the Agency for Cultural Affairs balance machine-learning exceptions with creator protections, and the government has said intellectual property policy must balance innovation incentives against rights-holder protection in creative industries. Japan's anime and gaming sectors, which feed both domestic models and global demand for AI-generated content, are watching how strictly those rules are applied in practice.

The divergence is visible in day-to-day use. Japanese creators who tested the same prompts across platforms describe Chinese tools as more permissive with stylised character likenesses, while US platforms routinely refuse requests that touch on copyrighted franchises. That flexibility has made Kling and Doubao popular shortcuts for short-form advertising and social clips in Japan, even as studios debate whether the shortcuts undercut the hand-drawn anime aesthetic that built the industry's global reputation.

Japan's 2026 guidance also carries a diplomatic dimension. The EU-Japan IP Action framework signed this year signals closer enforcement alignment with Europe, which has taken a harder line on AI training data than either the US or China. If Tokyo tilts toward Brussels, Japanese developers lose the permissive-training advantage that has helped Chinese rivals move fast; if it tilts toward pragmatism, it risks alienating the anime rights holders who are among the most vocal lobbyists in the creative economy.

Cheaper by the Minute: The Pricing War in AI Video

Open-weight Chinese developers are competing aggressively on cost. MiniMax's H3, which can generate 15-second 2K videos with stereo sound, is priced at one-third the level of mainstream competing products, according to the company, and has been made available through ByteDance's Dreamina creative platform for global creators.

Artificial Analysis estimates based on default API settings put the cost of generating one minute of 1080p video at about US$7.80 through MiniMax H3, compared with US$6 for Google's Gemini Omni Flash and US$9.07 through ByteDance's Seedance 2.0. Effective commercial prices for Chinese models are often discounted further through enterprise agreements, with Wan 3.0 and Seedance 2.0 trading at substantial markdowns on their headline rates.

Kuaishou's Kling Shows the Money Flowing Into AI Video

The technical and pricing edge is translating into top-line revenue. Kuaishou Technology said its Kling AI video-generation platform generated more than 850 million yuan (US$118 million) in second-quarter revenue, up more than 200 per cent year on year. The result is a rare bright spot for a company whose traditional live-streaming business has been squeezed by competition and user fatigue.

Kling's growth also shows how quickly AI video has moved from novelty to paid product. The platform now charges for subscriptions and usage tiers rather than giving away unlimited generation, and its enterprise clients range from e-commerce merchants producing product clips to film studios storyboarding scenes. Chinese vendors are effectively training global customers to expect high quality at a fraction of the cost of traditional production, a habit that will be hard to break once established.

By June, Kling AI's user base had topped 100 million globally, up about 67 per cent from late 2025, while its enterprise client roster expanded by roughly 67 per cent to nearly 50,000 businesses, according to Kuaishou. Kling 3.0's 1080p Pro version ranks ninth on the Artificial Analysis text-to-video leaderboard with audio.

What the AI Video Race Means for Japan

The Chinese advance poses both a competitive challenge and an opportunity for Japanese companies. Japanese studios and broadcasters experimenting with generative video face a choice between adopting low-cost Chinese platforms such as Kling, Wan and Seedance, or paying premiums for US or domestic alternatives under stricter copyright regimes.

Japan's own AI ambitions have concentrated on language models and physical AI, with a government-backed industry consortium announced in April bringing together SoftBank, NEC, Sony and Honda around a trillion-yen push. Video generation has received less strategic attention, leaving Japanese content firms reliant on foreign platforms for a technology that is rapidly reshaping the anime and advertising industries.

The copyright question cuts both ways. Japan's relatively permissive machine-learning provisions could give domestic developers room to train on Japanese content, but rights holders in the anime and manga industries have pushed for stronger protection, echoing the US debate that Chinese rivals have been able to sidestep.

What to Watch For

The immediate test is whether Chinese video models can convert benchmark leadership into durable global market share as US platforms respond. Watch for the next round of model releases from Google and OpenAI, and for how aggressively ByteDance and Alibaba price their enterprise offerings to lock in customers.

For Japan, the more consequential question is regulatory: how the JPO and the Agency for Cultural Affairs apply the new 2026 guidance to AI training on anime and other copyrighted works, and whether Tokyo moves to support a domestic video-generation ecosystem before its creative industries become dependent on Chinese platforms. The window for that decision, like the model race itself, is closing faster than most policymakers expect.

By Kenji Tanaka, Staff Writer

This article was produced with AI-assisted research and editorial support. Sources: South China Morning Post, Kuaishou Technology, Artificial Analysis, Japan Patent Office.