Alibaba’s RynnBrain, ByteDance’s Seedance 2.0 are among China’s new AI

While the U.S. market is focused on the impact Anthropic and Altruist’s tools will have on software and financial services, the Chinese tech giant this week released an AI model that shows advances in robotics and video generation.

alibabaTikTok creator ByteDance and short video platform Kuaishoreleased a new AI model highlighting how Chinese companies are catching up with their U.S. counterparts

This was announced after Demis Hassabis, the head of Google DeepMind, told CNBC that China’s AI models were only “a few months” behind their Western rivals.

These Chinese models are in direct competition with video generation models such as OpenAI’s Sora, as well as Chinese robotics models. Nvidia and google.

Here is an overview of the model.

Alibaba’s DAMO Academy this week unveiled RynnBrain, an AI model designed to help robots understand the physical world around them and identify objects.

In a video demonstration, Alibaba showed a robot with scissors in its hands counting oranges, picking them up and putting them in a basket. He was also shown taking out milk from the refrigerator.

Extensive training is required to enable the model to identify everyday objects with which it interacts. This means that simple tasks like picking up fruit can be difficult with robotics.

LinBrain currently puts Alibaba in competition with companies such as: Nvidia and google A company that develops unique AI models for robots.

“One of its key innovations is that it has built-in temporal and spatial awareness,” Hugging Face researcher Adina Yakev told CNBC.

“Rather than simply reacting to immediate inputs, robots can remember when and where events occur, track progress on tasks, and continue across multiple steps. This makes robots more reliable and consistent in complex real-world environments.”

Yakev added that Alibaba’s “broader ambition” is to “establish a foundational intelligence layer for the embodied system.”

Seedance 2.0 is a video generation AI model that can generate realistic videos from just text prompts from users. However, prompts can also include other videos and images.

A video created with Seedance 2.0 and reviewed by CNBC appears to show highly realistic images and videos created entirely with AI.

Billy Bowman, who runs a Stockholm, Sweden-based creative advertising agency that produces AI-generated content, uses Seedance 2.0.

He said AI video generation has made significant progress over the past two years, with rapid improvements seen across the industry.

“Back in 2023… it was difficult to make someone run or walk. Any realism was (limited to) very short clips, everything was very slow, bad textures, no skin texture, lacked detail. Now the script has been flipped. Now you can do anything. The advances in technology have been extraordinary,” Bowman said in an interview with CNBC.

Hugging Face’s Yakev added that the Seadance 2.0 model shows progress over the previous generation in “controllability, speed and production efficiency.”

“Seedance 2.0 is one of the most well-rounded video production models I’ve ever tested. I was really surprised at how satisfying the results were on the first try, even with simple prompts. The visuals, music, and cinematography come together in a way that feels sophisticated rather than experimental,” said Yakev.

But while users praised the technology, Seedance ran into problems. Local media in China reported that Seadance has suspended its ability to generate human voices based on the photos uploaded by its AI. This comes after Chinese bloggers raised concerns about audio being generated without their consent.

ByteDance did not immediately respond to a request for comment from CNBC.

Kuaishou’s Kling 3.0, released last week, is another video generation model that rivals ByteDance.

Kling 3.0 features significant consistency upgrades, photorealistic output, increased video playback time up to 15 seconds, and native audio generation across multiple languages, dialects, and accents.

Kuaishou said the model is only available to paying subscribers, but will be made available to the public soon.

Kuaishou’s success with the Kling model has been a major factor in the company’s stock price rising more than 50% in the last year.

Zhipu AI, trading in Hong Kong as Knowledge Atlas Technology, saw its stock price soar on Thursday after releasing GLM-5, an open source large-scale language model with enhanced coding capabilities and long-running agent tasks.

The company says the model approaches Anthropic’s Claude Opus 4.5 in coding benchmarks and beats Google’s Gemini 3 Pro in some tests. CNBC was unable to verify those claims.

MiniMax’s stock also soared on Thursday after announcing its latest M2.5 open source model with enhanced AI agent tools. “Agent” or “Agent AI” refers to an AI tool designed to automate tasks.

—CNBC’s Anniek Bao and Dylan Butts contributed to this report.