MiniCPM-V-2_6
openbmb/MiniCPM-V-2_6 · Apache-2.0
A 3B model that sees, hears and reads — with excellent OCR.
MiniCPM-V-2.6 is a small multimodal model strong at document and scene understanding, multilingual OCR and conversational image+video use, runnable on a laptop GPU.
Score
Setup at a glance
pip install transformers Deep dive
MiniCPM-V 2.6 is OpenBMB's 8B-class multimodal model that handles images, video and voice in one checkpoint, with OCR and reasoning as its standout strengths — competitive on document and chart understanding well above its size.
It supports multi-image input (compare several images in one conversation), video understanding, and runs as an image-text-to-text model in transformers with custom code from the repo.
It is gated on Hugging Face (accept the license first), and the model card documents the full capability matrix from OCR to feature extraction.