Ant Group officially released its trillion-dollar thinking model, Ring-1T, which was open-sourced upon release.

The Bailing team officially released the trillion-thinking model Ring-1T and announced its open-source status. Based on the Ling 2.0 architecture, the model boasts 1T total parameters and 50B activation parameters, supporting up to 128K context windows. Trained using large-scale verifiable reward reinforcement learning (RLVR), Ring-1T demonstrates outstanding performance on challenging tasks such as math competitions, code generation, and logical reasoning, achieving a leading level in open-source computing.