In the early morning of October 14, Ant Group released the trillion-parameter thinking model Ring-1T and fully open-sourced the model weights and training recipes. Based on the preview version, this model has enhanced its natural language reasoning capabilities through expanded RLVR training, while also improving its general capabilities using RLHF training, resulting in more balanced performance across various task benchmarks. This marks the first attempt by Ant Group's Bailing team in the field of trillion-parameter thinking models, with plans to continue optimizing performance in the future. Currently, users can download or experience the model online.