PANews reported on March 7 that according to Jinshi, Xinkaipu said in an institutional telephone exchange on March 6 that the company's self-developed Xingpu large model, through the training technology of SFT+RL (supervised fine-tuning + reinforcement learning), has achieved similar intelligent reasoning effects as DeepSeek-R1 in internal evaluation and verification, while the computing power consumption is only about 1/20 of that. The next step will be to participate in industry evaluation to obtain industry-recognized data. Reducing computing power consumption can reduce hardware investment, and the saved computing power budget can provide space for transfer to software, providing space for increasing budgets for more software, intelligent services and intelligent applications, while reducing computing power hardware investment while still ensuring user experience and service accuracy.