Model
Inference & Serving
MiniCPM-V 2.6
1GPT-4V-level 2multimodal 3model 4runs 5on 6phones
the six Ws · specification
W1
Who
OpenBMB, affiliated with Tsinghua NLP, developed the MiniCPM-V series.
W2
What
A compact 8-billion-parameter multimodal model designed for on-device image, video, and OCR understanding.
W3
Where
Published on Hugging Face and GitHub under the openbmb organization.
W4
When
MiniCPM-V 2.6 was released in August 2024.
W5
Why
It brought strong multimodal performance to edge devices like phones, unusual among competing VLMs.
W6
With
Runs via llama.cpp, Ollama, or transformers, and is optimized for mobile and iPad deployment.
W7
Watch
Open-weight release by OpenBMB under Apache 2.0 license.
alternatives
works with