Haha, deepseek r1 is using a modified BoN-RL replacing BoN with Group mean advantage was. And Kimi is taking the formulation of BoN it self. Amazing to see those model become life