Ant Lingbo Embodied Model Challenge Draws Record 1,855 Teams; Finals Set for Shanghai

The contest was launched in September at the 2026 Bund Conference by Ant Lingbo Technology with ModelScope and Alibaba Cloud Tianchi. It uses the open-source LingBot-VLA2.0 model as its technical base and focuses on model training and tuning, simulation generalization evaluation and real-robot adaptation. It was open to companies, universities, individual developers and technology enthusiasts worldwide.

Organizers said participants came from 648 universities, research institutes, leading technology companies and embodied-AI firms, including 41 overseas universities and organizations. The roster included teams from Tsinghua University, Peking University, Shanghai Jiao Tong University, Zhejiang University, the University of Science and Technology of China, Tongji University, the University of Chinese Academy of Sciences and Harbin Institute of Technology, as well as the University of Hong Kong, Hong Kong University of Science and Technology and the Chinese University of Hong Kong. International participants included researchers from Carnegie Mellon University, Georgia Institute of Technology, the University of Southern California, University College London, the Technical University of Munich, the National University of Singapore, Nanyang Technological University, the University of Sydney, KTH Royal Institute of Technology and Waseda University.

Industry developers also joined. Teams came from Huawei, Tencent, Alibaba, ByteDance, Meituan, Microsoft, BYD and Xiaomi, according to the report, while domestic humanoid-robot and embodied-AI companies including UBTech, Fourier Intelligence, AgiBot and the Beijing Humanoid Robot Innovation Center also took part.

The organizers’ profile showed that 74.7% of participants were from universities or research institutions, with master’s and doctoral students making up 67.4%. Developers from companies and frontline roles accounted for 25.2%. On technical background, 51.7% had experience with VLA or multimodal large models; 48% with deep-learning training; 31.9% with reinforcement learning; 25.4% with simulation environments; 23.4% with model deployment; 20.9% with real-robot debugging; and 20.1% with robot control.

According to the report, Ant Lingbo’s open-source models had more than 33,000 GitHub stars. LingBot-VLA and LingBot-VA had been adapted to LeRobot, a large robot development platform, and LingBot-Vision had been merged into the timm backbone library. LingBot-World was described as the first world model supported by both vLLM and SGLang, two mainstream open-source inference engines.

The offline final and the first Ant Lingbo Developer Conference will be held Nov. 14-16 in Shanghai. The 15 teams will compete in a 48-hour real-robot hackathon, with preliminary registration having closed at midnight on Oct. 27.