Beijing Humanoid Robot Innovation Center releases latest open-source VLM model.

Recently, the Beijing Humanoid Robot Innovation Center has fully open-sourced the Embodied Intelligence VLM model - Pelican-VL 1.0. According to the introduction, this model covers a parameter scale of 7B and 72B, making it the "largest open-source embodied multimodal large model" to date. According to tests, its performance surpasses similar models like GPT-5 by 15.79%, and also surpasses domestic models like Tongyi Qianwen and Shusheng Wanxiang, making it the strongest open-source embodied multimodal large model currently available. The open-sourcing of Pelican-VL 1.0 can significantly enhance the ability of embodied intelligence in various real scenarios such as commercial services, industrial pan-industry, high-risk special operations, and home services, through the visual-language perception to assist in multi-step task planning.