MiniCPM-V 2.6 is a multimodal large language model based on 800 million parameters, demonstrating leading performance in single image understanding, multiple image understanding, and video comprehension across various domains. The model achieved an average score of 65.2 on multiple popular benchmarks such as OpenCompass, surpassing widely used proprietary models. It possesses robust OCR capabilities, supports multiple languages, and performs efficiently, enabling real-time video understanding on devices like the iPad.