On August 26th, Wallface Intelligence unveiled its open-source 8B parameter MiniCPM-V 4.5 multi-modal model, distinguished by its impressive "high refresh rate" video understanding capabilities. The model is heralded for its industry-leading video understanding performance, surpassing the Qwen2.5-VL 72B model. At the same visual token cost, MiniCPM-V 4.5 can process six times the number of video frames compared to conventional models, achieving a remarkable 96-fold visual compression rate and operating with an efficiency that is 12-24 times higher than similar models.