MiniCPM-V 4.6

MiniCPM-V 4.6 is a 1.3B-parameter open-weight vision-language model from OpenBMB designed for edge and on-device deployment, capable of understanding images, multi-image inputs, and video alongside text. It uses mixed visual token compression to cut visual encoding compute by over 50%, enabling competitive multimodal performance on consumer and mobile hardware. Source: supabase.