Kuaishou (01024) launches end-to-end generative recommendation system OneRec large model technology reconstructs traditional recommendation architecture

Recently, Kuaishou (01024) launched a new end-to-end generative recommendation system, OneREC, to achieve a double breakthrough in effectiveness and cost. The system reconstructs the traditional recommended architecture through large model technology, increasing the effective calculation volume by 10 times, and the computing power utilization rate reaching 23.7% and 28.8%, and the operating cost is only 10.6% of the traditional solution. OneRec has been fully launched in the Kuaishou App and Express Edition, accepting approximately 25% of the QPS (number of requests per second), and the user stay time and life cycle have increased significantly. OneRec marks that the recommendation system has entered a new stage of “end-to-end generative awakening”.