Alibaba pushes into robotics AI with open-source ‘RynnBrain’

[BEIJING] Alibaba Group Holding debuted an AI model that can help robots and other devices perform real-world tasks, taking another step towards an eventual goal of leading multiple artificial intelligence spheres.

The Chinese company’s Damo Academy introduced an open-source foundation model that engages with the environment, understands space in relation to time and can figure out steps towards completing a task. Dubbed RynnBrain, the description suggests the model can help map objects, predict trajectories and navigate cluttered environments such as a kitchen or factory assembly line.

With the release, Alibaba takes on AI leaders including Alphabet’s Google and Nvidia. The Chinese firm claimed state-of-the-art results on benchmarks against Google’s Gemini Robotics-ER 1.5 and Nvidia’s Cosmos-Reason2.

Trained on Alibaba’s Qwen3-VL vision language model, RynnBrain is available on platforms like Hugging Face and GitHub in multiple versions starting as small as two billion parameters to a more efficient mixture-of-experts version.

It is designed partly for use in a field in which China and the US are vying for supremacy. Beijing has emphasised robotics, including humanoids, as a key prong of a strategic plan to dominate physical AI and reshape sectors from manufacturing to hospitality.

Chinese companies have largely focused on open-source AI, a contrast to the US approach of keeping cutting-edge technology behind proprietary walls.

SEE ALSO

China digital exports surge as Alibaba, Tencent lead global push

Alibaba to spend three billion yuan for Chinese New Year AI push as chatbot war heats up

Alibaba plans IPO for AI chipmaking unit T-Head: sources

Alibaba takes major step to link Taobao shopping to main AI app

Alibaba and JPMorgan are said to invest in chip designer Montage’s HK listing

More

DECODING ASIA

Navigate Asia in
a new global order

Get the insights delivered to your inbox.

The open sourcing strategy in physical AI – until now mainly confined to releases by academic institutions like Stanford University and University of California, Berkeley – could encourage developers and researchers around the world to refine the technology, eroding the West’s leadership. BLOOMBERG