Search
On November 13, 2025, the Beijing Humanoid Robot Innovation Center took a significant step by fully open-sourcing its cutting-edge embodied intelligence VLM model, named Pelican-VL1.0. This innovative model comes in two parameter scales: 7 billion (7B) and 72 billion (72B), positioning it as the largest open-source embodied multimodal large model currently available. To achieve its remarkable capabilities, the model underwent intensive training on a cluster comprising over 1,000 A800 GPUs. In terms of performance, it outperforms similar GPT-5 models by a notable margin of 15.79%. Furthermore, when compared to Google's renowned Gemini series models, Pelican-VL1.0 demonstrates a substantial improvement of 19.25%, showcasing its advanced technological prowess.
-
C114 Communication Network -
Communication Home
