MiniCPM-V 2.6
OpenBMB · 2024-08-06 · 8.1B parameters
MiniCPM-V 2.6 is OpenBMB's 8.1B-class vision-language model built on SigLIP-400M and Qwen2-7B, released in August 2024. It supports single-image, multi-image, and video understanding, processes images at any aspect ratio up to 1.8 million pixels, and was reported to outperform GPT-4V across single-image, multi-image, and video evaluations while supporting real-time video understanding on an iPad.